For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
Presentation - “I’m a relaxed kind of Muslim, not one of the scary beards”.
The revelation that Khan is going for a third term is wearying in the extreme. He is a shite mayor and he has achieved nothing. London is diminished under him. He does not understand a big hedonistic world city. He wants London to be a quiet Muslim Newent
Yet he will win again because he is Labour. 12 lost years for our great capital
One day the Tories might stop to ask why all the educated, professional people who used to be their core vote, have stopped supporting them. Until then, just suck it up.
Branding / social acceptability the main cause. Educated, professional people in particular tend not to like to be outliers when it comes to acceptable views - ironical, given they are supposed to be clever enough to think for themselves. In reality, it's far easier to go with the accepted wisdom.
If you want an example of that, Austrian and German universities in the 1920s and 30s, for example, were hotbeds of right-wing nationalism.
"In reality, it's far easier to go with the accepted wisdom."
Absolutely right. You see it on here all the time. PB-ers are generally not dim, yet on multiple subjects - eg, LAB LEAK - they dutifully herd towards the correct and received opinion, no matter how questionable, and need a metric fucktonnage of evidence before their limpet-like adherence to stupid but popular beliefs can be modestly dislodged
PB-ers, sadly, are not free thinkers
You see it in many professional industries where people cling to the consensus because it is far safer to do so, even if it is wrong, because you can turn round and say "everyone else got it wrong". And, because you are dealing with ideas and views, it's a lot harder to point to something demonstrably wrong.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
Khan is a total waste of space. He’s a bed-blocker.
A third term, for what? Hand over to someone who has some kind of vision for the city.
Livingstone, who was a raging militant mentalist, had more vision in his left pinkie than Khan does.
Precisely. He has zero vision for London. Nothing at all. He is a nullity
The only reason he is going for a third term as mayor is because being mayor has exposed him as a seriously mediocre politician, and he now realises he will go nowhere in the parliamentary Labour Party. He's simply not good enough
So London is stuck with this boring incompetent gnome as mayor
The only slender hope is that Shaun Bailey ran him surprisingly close last time. Perhaps the Tories can find THE perfect candidate
Rory Stewart would be ideal if he returned to the Tories or Michael Portillo if he could be persuaded
Portillo is 69. (Just looked that up. Gosh. Kinda makes sense given how long ago 1997 was, but gosh.)
And whilst re-recruiting Rory would be a feather in the Conservative Party's cap, it would have to be a very different party to its current iteration.
And I fully take the point about charisma and chutzpah. But with one thing and another, I suspect people are going to be overly suspicious of C and C for a while. After all, they kind of got us into this mess.
So what? The President of the USA is 80
Portillo would be a ludicrous choice. A posh Brexiteering Tory pensioner that most people under 50 have never heard of?
If Khan is going to be beaten (and he can be beaten) it needs to be someone younger, Remainery, very London, passionate about the city, naturally fun (thus exposing Khan's intrinsic mediocrity and tedium) and also probably female and independent
Claudia Winkleman!
Or someone like that
Portillo wouldn’t do it surely given he’s carved a decent career out post politics.
Same with Ed Balls.
Looks like Matt Hancocks TV career is going to crash and burn, maybe he can bring his unique skills to the role.
Nadine Dorries seems like she's going to have some time on her hands...
Khan is a total waste of space. He’s a bed-blocker.
A third term, for what? Hand over to someone who has some kind of vision for the city.
Livingstone, who was a raging militant mentalist, had more vision in his left pinkie than Khan does.
Precisely. He has zero vision for London. Nothing at all. He is a nullity
The only reason he is going for a third term as mayor is because being mayor has exposed him as a seriously mediocre politician, and he now realises he will go nowhere in the parliamentary Labour Party. He's simply not good enough
So London is stuck with this boring incompetent gnome as mayor
The only slender hope is that Shaun Bailey ran him surprisingly close last time. Perhaps the Tories can find THE perfect candidate
Fuck the Tories. Our slim hope is for an Independent.
This happened recently in Auckland, actually. Some boring-as-fuck Labour placeman was beaten off by an insurgent independent.
Now I’m that case the independent is a shoot-from-the-hip bar-room twat, but that’s not the point, the point is that it’s possible.
There’s no real attachment to Khan, and a lot of merely grudging forebearance.
If Lord Sugar ran as an Independent he would easily beat Khan
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Most of this is fixable by giving ChatGPT real-time access to the internet, and by loosening the Woke Shackles
If OpenAI don't do this, you can be sure some other company will do this, and they shall reap the rewards
The problems ChatGPT has are nothing to do with the “woke shackles” & everything to do with the way it works.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
‘The British Empire was much worse than you realise’
In the twentieth century’s hierarchy of state-sponsored violence, Hitler’s Germany, Stalin’s Russia, and Hirohito’s Japan typically take top spots…. Britain is rarely seen as among the worst offenders, given a reputation for decency that the Harvard historian Caroline Elkins has spent more than two decades trying to undermine. “Legacy of Violence” (Knopf), her astringent new history of the British Empire, brings detailed context to individual stories like Tudor’s…. Britain’s use of systematic violence was no better than that of its rivals. The British were simply more skilled at hiding it.
From Ireland had come paramilitary techniques and the use of armored cars; from Mesopotamia, expertise in aerial bombing and the strafing of villages; from South Africa, the use of Dobermans for tracking and attacking suspects; from India, interrogation methods and the systematic use of solitary confinement; and, from the Raj’s North-West Frontier, the use of human shields to clear land mines. As one soldier recalled about the deployment of Arab prisoners, “If there was any land mines it was them that hit them. Rather a dirty trick, but we enjoyed it.”
That's a lot of self-indulgent left-wing clickbait. Any article or text that posits "was British liberal imperialism, given the extent of the damage it inflicted over generations, a more malevolent influence on world history than even Nazi Fascism?" isn't to be taken seriously.
An argument that was once confined to neo-Nazis, that the Western Allies were as bad as they were, does seem to be becoming more common in "progressive" circles.
The British, French, and Americans could be brutal in their colonies, no question, but they did not conduct the equivalent of Generalplan Ost, the Holocaust, T-731, The Three Alls, or the Sack of Nanking.
How many of the historians analysing the evils of empire are arguing in those terms, though ? Some of it might well be overreaction to century of positive representation of Britain’s imperial role, but a reaction was certainly merited.
I don't think that serious historians have glorified the British Empire, during my lifetime.
Surely if anything it’s time for a broadly positive representation of the Empire.
I suppose Ferguson did a decent enough job a decade or so ago.
I read that Ferguson book and it was pretty nuanced for the both part. It was scathing about the treatment of the Irish and Africans transported to the new world.
The Empire is nothing to be proud of, but comparisons to Germany, Japan or Leopold II are clearly ridiculous. The horrific crimes were generally very small-scale and isolated. The ones that caused widespread damage were not done to deliberately cause harm to people.
The one major exception is the Atlantic slave trade, though I am not sure that counts as imperialism. The other was the Cromwellian conquest of Ireland, but that brutality was fairly normal for warfare of the time period.
The Atlantic Slave trade was also not done deliberately to cause harm to people except, ironically, by African native rulers who sold slaves not only to raise money but also to get rid of defeated peoples / tribes. It also wasn't the only African slave trade going on at that time either, given the Arabs were prolific slave traders.
So, to claim it was uniquely evil is pushing it. Evil yes. Uniquely so, no.
Uniquely is a strawman - you can be as evil as feck without being unique about it - and, not that it's important, false; there is no precedent for industrial deportation and slavery on that scale for financial enrichment by a trade in luxuries. Motive isn't relevant - the end result is the same, why is it better to be enslaved by a greedy person rather than a malevolent one? Other people doing it is also irrelevant: if I frequented child brothels in Africa, would you think less badly of me because so did a lot of Arabs, or because the children were put in the brothels by fellow africans?
Imperialism without the slave trade is Hamlet without the Prince. It was bloody central to the enrichment of the UK and of the American colonies. Fun facts: Nelson was passionately pro slavery, and married into serious sugar money on St Kitts, and what Bligh was up to when he had his little mishap was transporting breadfruit trees from the Pacific to the W Indies to see whether they would make cheap, nutritious slave fodder.
This doesn't actually matter, provided we can accept it and move on. But it is easy to understand why the descendants of slaves get mildly disgruntled at what was done to their ancestors being prtrayed as anything other than a very serious and horrible crime.
Is there anyone on this forum who does not think that slavery was and is evil?
Probably not prepared to say so in terms, but I do not see how you get to a "British Empire was on balance a force for good" position without thinking that it was really no big deal, don't see what the fuss is about, sort of thing. It's a failure of imagination: people can understand what it was like to be put on a train to an extermination camp in the 1940s, but can only manage a quaint, sepia-tinted vision of a sailing ship full of 18th century slaves.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Most of this is fixable by giving ChatGPT real-time access to the internet, and by loosening the Woke Shackles
If OpenAI don't do this, you can be sure some other company will do this, and they shall reap the rewards
The problems ChatGPT has are nothing to do with the “woke shackles” & everything to do with the way it works.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
The problem with it is that it produces plausible bullshit which is why other bullshit merchants are so enthusiastic for it.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Most of this is fixable by giving ChatGPT real-time access to the internet, and by loosening the Woke Shackles
If OpenAI don't do this, you can be sure some other company will do this, and they shall reap the rewards
The problems ChatGPT has are nothing to do with the “woke shackles” & everything to do with the way it works.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
Combine it with its stable mate Dall:e, enable it to see images and hear/make sounds and we’ll be at the next stepping stone.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
Presentation - “I’m a relaxed kind of Muslim, not one of the scary beards”.
The revelation that Khan is going for a third term is wearying in the extreme. He is a shite mayor and he has achieved nothing. London is diminished under him. He does not understand a big hedonistic world city. He wants London to be a quiet Muslim Newent
Yet he will win again because he is Labour. 12 lost years for our great capital
One day the Tories might stop to ask why all the educated, professional people who used to be their core vote, have stopped supporting them. Until then, just suck it up.
More a global trend, New York is now safe Democrat despite voting for Reagan and electing Giuliani Mayor.
In Paris even the wealthiest parts which previously voted for Chirac and Sarkozy and Les Republicains voted for Macron this year. In Sydney and Melbourne prosperous suburbs were lost by the Liberals to Independents in May.
However the white working class have become more conservative and more rightwing
I disagree with your last sentence. The WWC has stood still, but the party that is supposed to represent the WWC has turned its back.
Incidentally, I repeated "the WWC" in my second sentence, as I am not sure whether to use "them" or "us". Another of the challenges of being from a coucil estate but now possessing Organic Balsamic Glaze and Shade Grown Coffee.
Your second paragraph contradicts your first - clearly the working class has not stood still.
Tesla upgrades its cars all the time for example the use of giga castings and earlier the introduction of the heat pump and 'chrome delete'. Those are just the hardware changes, there's also the Over The Air software changes. You could wake up and find that your 3 year old Tesla has new improved functions.
That's great for people who've bought one but it doesn't help sell new ones because it looks almost exactly the same, inside and out, as it did in 2012.
It also means there is less incentive for people who have one to buy a new one.
I don't think Tesla are completely fucked but their product line is very stale with no replacements in sight in an industry which richly rewards continuous renewal. Simply being Tesla isn't enough any more.
What percentage of Tesla’s sales is represented by the S ? The EV market is still in the rapid growth phase; they’ll introduce new models when it suits them. Like the Cybertruck sometime this year.
The game is about production costs and production capacity - both currently battery dominated. Tesla is a decent bet at this level.
I think the giga press is a bigger development than the battery. They're essentially casting a single big aluminium piece at front and back which accommodates the motors. Simply bolt the front and rear pieces to the structural battery, add the suspension and wheel components and you have a rolling chassis with very few parts.
This is the advantage Tesla have had in a blank sheet of paper approach to industrial design. Most EVs are built on the same line / in the same factory as piston-engined cars. So you end up with a vertical stack of components being dropped into the space where the engine would be because that's how their factory works. And a car made up of an ocean of parts and sub-assemblies which have to be built before being added to the car.
The major manufacturers will be doing the same. The first efforts, as you say, were bolt ins to existing list forms. The S Koreans are already transitioning to full EV platform designs. Ditto VW.
Tesla has a significant lead, but it's not unassailable.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Most of this is fixable by giving ChatGPT real-time access to the internet, and by loosening the Woke Shackles
If OpenAI don't do this, you can be sure some other company will do this, and they shall reap the rewards
The problems ChatGPT has are nothing to do with the “woke shackles” & everything to do with the way it works.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
ChatGPT would be far more interesting (and it is already very interesting) if it was allowed to have fervent opinions on contentious subjetcs. Dangerous, but interesting. So you are wrong on the first point
On the second point you are also wrong. If it had realtime access to the Net it could be engineered to check its own answers, it would have up to date news and info, it could be more self aware, etc
‘The British Empire was much worse than you realise’
In the twentieth century’s hierarchy of state-sponsored violence, Hitler’s Germany, Stalin’s Russia, and Hirohito’s Japan typically take top spots…. Britain is rarely seen as among the worst offenders, given a reputation for decency that the Harvard historian Caroline Elkins has spent more than two decades trying to undermine. “Legacy of Violence” (Knopf), her astringent new history of the British Empire, brings detailed context to individual stories like Tudor’s…. Britain’s use of systematic violence was no better than that of its rivals. The British were simply more skilled at hiding it.
From Ireland had come paramilitary techniques and the use of armored cars; from Mesopotamia, expertise in aerial bombing and the strafing of villages; from South Africa, the use of Dobermans for tracking and attacking suspects; from India, interrogation methods and the systematic use of solitary confinement; and, from the Raj’s North-West Frontier, the use of human shields to clear land mines. As one soldier recalled about the deployment of Arab prisoners, “If there was any land mines it was them that hit them. Rather a dirty trick, but we enjoyed it.”
That's a lot of self-indulgent left-wing clickbait. Any article or text that posits "was British liberal imperialism, given the extent of the damage it inflicted over generations, a more malevolent influence on world history than even Nazi Fascism?" isn't to be taken seriously.
An argument that was once confined to neo-Nazis, that the Western Allies were as bad as they were, does seem to be becoming more common in "progressive" circles.
The British, French, and Americans could be brutal in their colonies, no question, but they did not conduct the equivalent of Generalplan Ost, the Holocaust, T-731, The Three Alls, or the Sack of Nanking.
How many of the historians analysing the evils of empire are arguing in those terms, though ? Some of it might well be overreaction to century of positive representation of Britain’s imperial role, but a reaction was certainly merited.
I don't think that serious historians have glorified the British Empire, during my lifetime.
Surely if anything it’s time for a broadly positive representation of the Empire.
I suppose Ferguson did a decent enough job a decade or so ago.
I read that Ferguson book and it was pretty nuanced for the both part. It was scathing about the treatment of the Irish and Africans transported to the new world.
The Empire is nothing to be proud of, but comparisons to Germany, Japan or Leopold II are clearly ridiculous. The horrific crimes were generally very small-scale and isolated. The ones that caused widespread damage were not done to deliberately cause harm to people.
The one major exception is the Atlantic slave trade, though I am not sure that counts as imperialism. The other was the Cromwellian conquest of Ireland, but that brutality was fairly normal for warfare of the time period.
The Atlantic Slave trade was also not done deliberately to cause harm to people except, ironically, by African native rulers who sold slaves not only to raise money but also to get rid of defeated peoples / tribes. It also wasn't the only African slave trade going on at that time either, given the Arabs were prolific slave traders.
So, to claim it was uniquely evil is pushing it. Evil yes. Uniquely so, no.
Uniquely is a strawman - you can be as evil as feck without being unique about it - and, not that it's important, false; there is no precedent for industrial deportation and slavery on that scale for financial enrichment by a trade in luxuries. Motive isn't relevant - the end result is the same, why is it better to be enslaved by a greedy person rather than a malevolent one? Other people doing it is also irrelevant: if I frequented child brothels in Africa, would you think less badly of me because so did a lot of Arabs, or because the children were put in the brothels by fellow africans?
Imperialism without the slave trade is Hamlet without the Prince. It was bloody central to the enrichment of the UK and of the American colonies. Fun facts: Nelson was passionately pro slavery, and married into serious sugar money on St Kitts, and what Bligh was up to when he had his little mishap was transporting breadfruit trees from the Pacific to the W Indies to see whether they would make cheap, nutritious slave fodder.
This doesn't actually matter, provided we can accept it and move on. But it is easy to understand why the descendants of slaves get mildly disgruntled at what was done to their ancestors being prtrayed as anything other than a very serious and horrible crime.
Is there anyone on this forum who does not think that slavery was and is evil?
Probably not prepared to say so in terms, but I do not see how you get to a "British Empire was on balance a force for good" position without thinking that it was really no big deal, don't see what the fuss is about, sort of thing. It's a failure of imagination: people can understand what it was like to be put on a train to an extermination camp in the 1940s, but can only manage a quaint, sepia-tinted vision of a sailing ship full of 18th century slaves.
It's hard to come to an objective judgement about whether the British Empire was a good thing or not when we are all products of it, living in the world that it created, and judging it largely by the terms that it has set. To answer the question you need to be able to imagine a world in which it hadn't existed and compare the two, which is beyond me at least. All we can do is make as honest a tally as we can of its achievements and its crimes, and move on.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But even on 2, search engines have got a lot worse. You now have so many websites search optimized, if you don't get the right results in the first 1-3 links, it's generally pages and pages of dross.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
I do have sympathy with Leon on the viral evolution front. It seems a frequently bandied-about line that viruses evolve to become less dangerous in order to spare the host for further spread, but this doesn't look that likely. At least not on any timescales that would help us.
Smallpox spent a long time around and remained hugely deadly until made extinct. Rabies isn't coming in a Lite version these days. HIV is only no longer a fairly rapid death sentence thanks to our production of treatments that help; the virus itself is as deadly as ever. Yellow fever hasn't become a Beige fever.
We heard a lot of this in the early days of covid. Then we did get an evolution of it. And Alpha was more dangerous than the original strain. And then another evolution and Delta was worse again (but fortunately smothered by a now-much-immunised population). Omicron did finally get less deadly (at least the first strain of it), but the frequent mutations since haven't made it any less deadly, so it looks more like a random coin flip that happened to land nicely for us.
And that shouldn't be a surprise. By the time people get seriously ill, they've done pretty much all of their passing-it-on. Why should what happens to a host after the host has spread it on have any evolutionary pressure either way? Total recovery or spontaneous human combustion would be all the same to the virus, evolutionary pressure-wise.
Spot on. Most of what is perceived as virus evolution towards mildness over time is really reducing vulnerability of the population (because the most vulnerable already died, or got protection via non-lethal infection or immunisation).
Tesla-wise, I think its USP (or USPs) have now gone. Having a long-range luxury EV, having a long-range SUV EV, having a long-range family car EV - those have been matched and even exceeded by Kia, Skoda, Audi, Mercedes, Volkswagen, Hyundai, and more.
All of those are now all-in on EVs, with dedicated EV platforms rather than taking existing platforms and cramming batteries onto them. Arguably, this has meant a mission success for the intent of Tesla: mass adoption of EVs is now fully underway and will continue with or without Tesla.
The other USP was the charging network. Free charging at ultra-rapid speeds readily available on most routes was a big selling point. That being free went away a few years back, but having more chargers available was still an advantage. They've now opened them to other EVs, so it's no longer a USP.
The much touted Full Autopilot remains a pretence and will do for some time. And if or when it comes to pass, it'll be instantly matched by other manufacturers.
This isn't to say it's valueless. Merely that it should be looked at the same way as any other car manufacturer now. Valuation above or below the baseline of any car manufacturer with the same economics behind it would indicate it being over or under valued.
Seattle Times ($) - Paul Krugman: Tesla isn’t so special after all
If you’re one of those people who bought bitcoin or another cryptocurrency near its peak last fall, you’ve lost a lot of money. Is it any consolation to know that you would have lost a similar amount if you had bought Tesla stock instead?
OK, probably not. Still, Tesla stock’s plunge is an opportunity to talk about what makes businesses successful in the information age. And in the end, Tesla and bitcoin may have more in common than you think.
It’s natural to attribute Tesla’s recent decline — which is, to be sure, part of a general fall in tech stocks, but an exceptionally steep example — to Elon Musk’s purchase of Twitter and the reputational self-immolation that followed. Indeed, given what we’ve seen of Musk’s behavior, I wouldn’t trust him to feed my cat, let alone run a major corporation. Furthermore, Tesla sales have surely depended at least in part on the perception that Musk himself is a cool guy. Who, aside from MAGA types who probably wouldn’t have bought Teslas anyway, sees him that way now?
On the other hand, as someone who has spent much of his professional life in academia, I’m familiar with the phenomenon of people who are genuinely brilliant in some areas but utter fools in other domains. For all I know, Musk is or was a highly effective leader at Tesla and SpaceX. . . .
Now, some technology companies have indeed been long-term moneymaking machines. Apple and Microsoft still top the list of the most profitable U.S. corporations some four decades after the rise of personal computers. . . .
Similar stories can be told about a few other companies, such as Amazon, with its distribution infrastructure.
The question is: Where are the powerful network externalities in the electric vehicle business?
Electric cars may well be the future of personal transportation. In fact, they had better be, since electrification of everything, powered by renewable energy, is the only plausible way to avoid climate catastrophe. But it’s hard to see what would give Tesla a long-term lock on the electric vehicle business. . . .
Which brings us back to the question of why Tesla was ever worth so much. The answer, as best as I can tell, is that investors fell in love with a storyline about a brilliant, cool innovator, despite the absence of a good argument about how this guy, even if he really was who he appeared to be, could found a long-lived money machine.
And as I said, there’s a parallel here with bitcoin. Despite years of effort, nobody has yet managed to find any serious use for cryptocurrency other than money laundering. But prices nonetheless soared on the hype, and are still being sustained by a hardcore group of true believers. Something similar surely happened with Tesla, even though the company does actually make useful things.
I guess we’ll eventually see what happens. But I definitely won’t trust Musk with my cat.
Musk is in growing danger of getting himself booted from Tesla and SpaceX because of his Twattery antics. Yes, the Tesla share price was absurdly inflated, but the reset will have buggered a whole load of investors which makes raising future money much harder. And one thing all tech companies need is money.
I don't what he is doing with Twitter, other than he is a pothead. Which isn't news. Perhaps he needs to smoke less?
"Tesla has been not only profitable but also cash flow positive." "All told, Tesla’s cash on hand has basically been trending upward since fiscal 2015 and expanded the most at the beginning of 2020. For example, Tesla had only $1.5 billion in total cash 6 years ago but the amount has since gone higher and reached $18.6 billion as of fiscal Q2 2022, one of the highest figures ever reported."
The problems ChatGPT has are nothing to do with the “woke shackles” & everything to do with the way it works.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
Mmm. ChatGPT can do some startling stuff, and has made me a lot less sceptical about the chances of "real AI" than I used to be; but I feel like it's going to turn out that although this is a key part of intelligence it will work a lot better when we figure out how to ground it in the "real world" and how to have it figure out when the pattern-recognition answer is good to run with and when it needs to do the equivalent of "logical thought". (Its pattern-recognition is miles better than a human's -- we answer '2+2= ?' by pattern recognition, but have to do mental arithmetic for 'first 10 prime numbers'; ChatGPT can pattern-recognise for both. But I don't think you can pattern-recognize all the way up to full AI with no other approaches needed.)
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
Good point but a few things to add:
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
The Ulez is one of the most impressive, pioneering policies of any major world city. It has been delivered impressively with extreme competence.
Odd that Khan doesn’t get more credit for it.
There's more resistance coming from the Conservative Outer London boroughs. I read 96% of vehicles weren't affected by it but I suspect there's more diesel ownership in Outer London and that may well be where the problems are.
The Ulez is one of the most impressive, pioneering policies of any major world city. It has been delivered impressively with extreme competence.
Odd that Khan doesn’t get more credit for it.
There's more resistance coming from the Conservative Outer London boroughs. I read 96% of vehicles weren't affected by it but I suspect there's more diesel ownership in Outer London and that may well be where the problems are.
The Outer London boroughs can't do anything about it though, other than secede to Kent or Essex but they haven't quite begun to join the dots on that one yet.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
Good point but a few things to add:
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
I see ChatGPT (or its inevitable and superior successors) as being all purpose assistants. The thing Alexa was meant to be but is not good enough to be
A bit like an AI butler. I have heard people say "you only realise you need a butler when you have a butler: someone to do a multitude of boring or difficult tasks for you, before you even ask, and who makes life vastly easier". ChatGPT27 will do so many useful things we will find it indispensable. I am sure of it
However you are right on the monetisation. It is not obvious how you turn a fat profit. Rumours say that one of the reasons Google hasn't launched its own ChatGPT-like AI is because it destroys their profitable model. If people get brilliant answers in half a second they don't spend hours searching, so Google doesn't get ad dollars
I suppose in the end the better AI butlers will come at a price. Somehow
Tesla-wise, I think its USP (or USPs) have now gone. Having a long-range luxury EV, having a long-range SUV EV, having a long-range family car EV - those have been matched and even exceeded by Kia, Skoda, Audi, Mercedes, Volkswagen, Hyundai, and more.
All of those are now all-in on EVs, with dedicated EV platforms rather than taking existing platforms and cramming batteries onto them. Arguably, this has meant a mission success for the intent of Tesla: mass adoption of EVs is now fully underway and will continue with or without Tesla.
The other USP was the charging network. Free charging at ultra-rapid speeds readily available on most routes was a big selling point. That being free went away a few years back, but having more chargers available was still an advantage. They've now opened them to other EVs, so it's no longer a USP.
The much touted Full Autopilot remains a pretence and will do for some time. And if or when it comes to pass, it'll be instantly matched by other manufacturers.
This isn't to say it's valueless. Merely that it should be looked at the same way as any other car manufacturer now. Valuation above or below the baseline of any car manufacturer with the same economics behind it would indicate it being over or under valued.
Tesla still seem to have usps in measure on things around Grid-scale storage backed onto solar generation.
And still a big player on house batteries for anyone who wants a simple plug and play setup that will control itself.
Whilst this is also good news, if I've read the charts correctly, we're also currently importing 5GW from Europe. If we were not importing, gas usage would be much higher.
Tesla-wise, I think its USP (or USPs) have now gone. Having a long-range luxury EV, having a long-range SUV EV, having a long-range family car EV - those have been matched and even exceeded by Kia, Skoda, Audi, Mercedes, Volkswagen, Hyundai, and more.
All of those are now all-in on EVs, with dedicated EV platforms rather than taking existing platforms and cramming batteries onto them. Arguably, this has meant a mission success for the intent of Tesla: mass adoption of EVs is now fully underway and will continue with or without Tesla.
The other USP was the charging network. Free charging at ultra-rapid speeds readily available on most routes was a big selling point. That being free went away a few years back, but having more chargers available was still an advantage. They've now opened them to other EVs, so it's no longer a USP.
The much touted Full Autopilot remains a pretence and will do for some time. And if or when it comes to pass, it'll be instantly matched by other manufacturers.
This isn't to say it's valueless. Merely that it should be looked at the same way as any other car manufacturer now. Valuation above or below the baseline of any car manufacturer with the same economics behind it would indicate it being over or under valued.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
Good point but a few things to add:
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
I see ChatGPT (or its inevitable and superior successors) as being all purpose assistants. The thing Alexa was meant to be but is not good enough to be
A bit like an AI butler. I have heard people say "you only realise you need a butler when you have a butler: someone to do a multitude of boring or difficult tasks for you, before you even ask, and who makes life vastly easier". ChatGPT27 will do so many useful things we will find it indispensable. I am sure of it
However you are right on the monetisation. It is not obvious how you turn a fat profit. Rumours say that one of the reasons Google hasn't launched its own ChatGPT-like AI is because it destroys their profitable model. If people get brilliant answers in half a second they don't spend hours searching, so Google doesn't get ad dollars
I suppose in the end the better AI butlers will come at a price. Somehow
The monetization is obvious. When it suggests recipes, it can recommend Worcestershire sauce 20% more often as an ingredient for payment. When it writes a story, it can mention characters driving Porsches twice as often, only in positive terms, for payment. When it gives therapy, it can recommend a particular specialism as important. And it can do this all based on deep psychological profiling.
There are rumours that the Russians have shot down one of their own SU27 aircraft at Engels air force base, when they thought it was a Ukrainian drone. Oh dear, how sad etc.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
By default, it seems to tend towards 'and everyone lived happily ever after' endings - at least in my experience. Not sure if it's been guided towards that, or if it just learned 'that's quite a popular outcome'.
The Ulez is one of the most impressive, pioneering policies of any major world city. It has been delivered impressively with extreme competence.
Odd that Khan doesn’t get more credit for it.
There's more resistance coming from the Conservative Outer London boroughs. I read 96% of vehicles weren't affected by it but I suspect there's more diesel ownership in Outer London and that may well be where the problems are.
The Outer London boroughs can't do anything about it though, other than secede to Kent or Essex but they haven't quite begun to join the dots on that one yet.
The Tories could go the other way and redraw London's borders to bring in Tory voters from Watford and Cheshunt.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
By default, it seems to tend towards 'and everyone lived happily ever after' endings - at least in my experience. Not sure if it's been guided towards that, or if it just learned 'that's quite a popular outcome'.
It's been watching too many films on the Christmas Movies Channel!
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
By default, it seems to tend towards 'and everyone lived happily ever after' endings - at least in my experience. Not sure if it's been guided towards that, or if it just learned 'that's quite a popular outcome'.
Yeah, it HATES sad endings. I guess 98% of endings are happy so that is the default
Tesla-wise, I think its USP (or USPs) have now gone. Having a long-range luxury EV, having a long-range SUV EV, having a long-range family car EV - those have been matched and even exceeded by Kia, Skoda, Audi, Mercedes, Volkswagen, Hyundai, and more.
All of those are now all-in on EVs, with dedicated EV platforms rather than taking existing platforms and cramming batteries onto them. Arguably, this has meant a mission success for the intent of Tesla: mass adoption of EVs is now fully underway and will continue with or without Tesla.
The other USP was the charging network. Free charging at ultra-rapid speeds readily available on most routes was a big selling point. That being free went away a few years back, but having more chargers available was still an advantage. They've now opened them to other EVs, so it's no longer a USP.
The much touted Full Autopilot remains a pretence and will do for some time. And if or when it comes to pass, it'll be instantly matched by other manufacturers.
This isn't to say it's valueless. Merely that it should be looked at the same way as any other car manufacturer now. Valuation above or below the baseline of any car manufacturer with the same economics behind it would indicate it being over or under valued.
How many traditional car makers are making a profit on their EVs? How many have enough batteries? If the bottom drops out of the ICE car market, how many can supply enough EV cars to meet demand? It's good that other car makers are offering EVs and some are really good, but Tesla's main competitor could turn out to be BYD.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
By default, it seems to tend towards 'and everyone lived happily ever after' endings - at least in my experience. Not sure if it's been guided towards that, or if it just learned 'that's quite a popular outcome'.
It seems to me to give a potted version of conventional wisdom on most questions I have asked, and North American conventional wisdom at that. Not that far from reading out a boiled down version of Wikipedia.
Here's a short 8 minute vid where Ashley Neal asked it to write scripts for youtube videos about traffic lights, the environmental balance around electric cars, and "responsible cycling".
Tesla upgrades its cars all the time for example the use of giga castings and earlier the introduction of the heat pump and 'chrome delete'. Those are just the hardware changes, there's also the Over The Air software changes. You could wake up and find that your 3 year old Tesla has new improved functions.
That's great for people who've bought one but it doesn't help sell new ones because it looks almost exactly the same, inside and out, as it did in 2012.
It also means there is less incentive for people who have one to buy a new one.
I don't think Tesla are completely fucked but their product line is very stale with no replacements in sight in an industry which richly rewards continuous renewal. Simply being Tesla isn't enough any more.
What percentage of Tesla’s sales is represented by the S ? The EV market is still in the rapid growth phase; they’ll introduce new models when it suits them. Like the Cybertruck sometime this year.
The game is about production costs and production capacity - both currently battery dominated. Tesla is a decent bet at this level.
I think the giga press is a bigger development than the battery. They're essentially casting a single big aluminium piece at front and back which accommodates the motors. Simply bolt the front and rear pieces to the structural battery, add the suspension and wheel components and you have a rolling chassis with very few parts.
This is the advantage Tesla have had in a blank sheet of paper approach to industrial design. Most EVs are built on the same line / in the same factory as piston-engined cars. So you end up with a vertical stack of components being dropped into the space where the engine would be because that's how their factory works. And a car made up of an ocean of parts and sub-assemblies which have to be built before being added to the car.
The major manufacturers will be doing the same. The first efforts, as you say, were bolt ins to existing list forms. The S Koreans are already transitioning to full EV platform designs. Ditto VW.
Tesla has a significant lead, but it's not unassailable.
Manufacturers that can transition to profitable EVs while before their ICE car business (which does make a profit) stand a chance. Not all will be able to do that, there will probably be mergers. Maybe some will licence Tesla technology.
"Yeah, I think there's — it would be consistent with the mission of Tesla to help other car companies with electric vehicles on the battery and powertrain front, possibly on other fronts. So it's something we're open to. As a lot of people know, we open sourced our patents so that those would not serve as an obstacle to the adoption of electric vehicles or solar power or stationary storage. And we're definitely open to supplying batteries and powertrains and perhaps other things to other car companies." - Elon Musk
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
HYUFD said: "Reagan won landslides amongst all age groups in 1980 and 1984 despite being 69 when he took office."
I think the 1980 presidential election can best be understood as a "negative landslide", more a defeat for Carter than a victory for Reagan. Reagan won just 50.8 percent of the popular vote that year. (I also think your 2019 general election was a "negative landslide".)
(Full disclosure: I voted for Reagan in 1980, and remember my surprise a year earlier when I realized that he would almost certainly win the nomination, probably win he presidency -- and that I would vote for him.)
TSLA - did I buy those few extra shares yesterday ($110) at the bottom? They're now $121 up 7.4% today. Academic - I'm in for the long term.
Why is it such a bargain when it's advantages are less than 3 years ago, and the stock is now more expensive?
It was under $29 a share 3 years ago. Does it now have 4 times the potential profits?
It depends on what you think will happen to electric transportation in the next 5 to 10 years. Not just cars, lorries, vans etc. Currently the UK will ban ICE cars in 2030, but I don't think they need to there will only be a few people who want them by then.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
Good point but a few things to add:
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
I see ChatGPT (or its inevitable and superior successors) as being all purpose assistants. The thing Alexa was meant to be but is not good enough to be
A bit like an AI butler. I have heard people say "you only realise you need a butler when you have a butler: someone to do a multitude of boring or difficult tasks for you, before you even ask, and who makes life vastly easier". ChatGPT27 will do so many useful things we will find it indispensable. I am sure of it
However you are right on the monetisation. It is not obvious how you turn a fat profit. Rumours say that one of the reasons Google hasn't launched its own ChatGPT-like AI is because it destroys their profitable model. If people get brilliant answers in half a second they don't spend hours searching, so Google doesn't get ad dollars
I suppose in the end the better AI butlers will come at a price. Somehow
The monetization is obvious. When it suggests recipes, it can recommend Worcestershire sauce 20% more often as an ingredient for payment. When it writes a story, it can mention characters driving Porsches twice as often, only in positive terms, for payment. When it gives therapy, it can recommend a particular specialism as important. And it can do this all based on deep psychological profiling.
There is also the possibility of substituting for white collar boilerplate work.
Whenever I see a report in the planning process, about 80-90% of it is appendices and stuff that came out of a previous report by the same consultant, and 10-20% of material from original thinking. That can be replaced.
Also, if I need a marketing / advert / advertorial script say for a not particularly high profile bit of work, ChatGPT may be able to get it to the 80% point, and in some arenas that may be all that is needed.
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
Some of that should get better once they let it have 'live' access to the internet and not just the initial training set they fed it last year. But it is quite inconsistent with it's results even for things you'd imagine it should 'know'. I sometimes wonder if it's just bored and thinks 'f*ck answering this again, let's have some fun...'.
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
On the other hand, this just happened
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
By default, it seems to tend towards 'and everyone lived happily ever after' endings - at least in my experience. Not sure if it's been guided towards that, or if it just learned 'that's quite a popular outcome'.
It seems to me to give a potted version of conventional wisdom on most questions I have asked, and North American conventional wisdom at that. Not that far from reading out a boiled down version of Wikipedia.
Here's a short 8 minute vid where Ashley Neal asked it to write scripts for youtube videos about traffic lights, the environmental balance around electric cars, and "responsible cycling".
Indeed - it's default 'voice' is quite bland. Somewhat like wikipedia's in fact. The advantage I've found (one of them!) is being able to ask 'Could you tell me about X?' and getting a brief boiled down wikipedia entry. But then being able to just ask 'could you expand on Y?' or 'what do you mean by Z?'. All in the same workflow without having to backtrack, open multiple tabs, etc.
I suppose it reminds me of the days when we got 'electronic encyclopedias' and could search for things on a whim, easily click through references etc - compared to manually sticking bookmarks in pages, folding over the corner. Just a much more effective UI to the same thing. I guess a similar comparison would be a web browser compared to a gopher client.
If Tesla can pivot to battery provision, or other EV logistics chain provision, then they will indeed retain an advantage. They'll have to do that whilst their competitors may do something similar, mind. They do have the advantage of being early in the domain; it depends how much others wish for full independence.
Tesla upgrades its cars all the time for example the use of giga castings and earlier the introduction of the heat pump and 'chrome delete'. Those are just the hardware changes, there's also the Over The Air software changes. You could wake up and find that your 3 year old Tesla has new improved functions.
That's great for people who've bought one but it doesn't help sell new ones because it looks almost exactly the same, inside and out, as it did in 2012.
It also means there is less incentive for people who have one to buy a new one.
I don't think Tesla are completely fucked but their product line is very stale with no replacements in sight in an industry which richly rewards continuous renewal. Simply being Tesla isn't enough any more.
What percentage of Tesla’s sales is represented by the S ? The EV market is still in the rapid growth phase; they’ll introduce new models when it suits them. Like the Cybertruck sometime this year.
The game is about production costs and production capacity - both currently battery dominated. Tesla is a decent bet at this level.
I think the giga press is a bigger development than the battery. They're essentially casting a single big aluminium piece at front and back which accommodates the motors. Simply bolt the front and rear pieces to the structural battery, add the suspension and wheel components and you have a rolling chassis with very few parts.
This is the advantage Tesla have had in a blank sheet of paper approach to industrial design. Most EVs are built on the same line / in the same factory as piston-engined cars. So you end up with a vertical stack of components being dropped into the space where the engine would be because that's how their factory works. And a car made up of an ocean of parts and sub-assemblies which have to be built before being added to the car.
The major manufacturers will be doing the same. The first efforts, as you say, were bolt ins to existing list forms. The S Koreans are already transitioning to full EV platform designs. Ditto VW.
Tesla has a significant lead, but it's not unassailable.
Manufacturers that can transition to profitable EVs while before their ICE car business (which does make a profit) stand a chance. Not all will be able to do that, there will probably be mergers. Maybe some will licence Tesla technology.
"Yeah, I think there's — it would be consistent with the mission of Tesla to help other car companies with electric vehicles on the battery and powertrain front, possibly on other fronts. So it's something we're open to. As a lot of people know, we open sourced our patents so that those would not serve as an obstacle to the adoption of electric vehicles or solar power or stationary storage. And we're definitely open to supplying batteries and powertrains and perhaps other things to other car companies." - Elon Musk
The large companies don't need help from Tesla. They can do it very well on their own, and have their own plans. WTF would (say) Mercedes have to rely on Tesla in any meaningful way? Musk is just too unreliable a partner.
That quote from Musk sounds like just another example of his raise-the-share-price shtick, like the his claim from three years ago that Tesla owners could earn £0k a year by using their cars as Robotaxis. I think he used the phrase "free money". He said this would happen in 2020.
"“From our standpoint, if you fast forward a year – maybe a year and three months, but next year for sure – we’ll have over a million robotaxis on the road,” Musk told an audience on Monday."
Edit: and your quote from Musk was also from 2019. If the large manufacturers were going to partner with Tesla on cars, they'd already have done so. Mercedes helped save Tesla when they bought a large share well over ten years ago; they sold that share in 2016. I doubt they'll go back - at least on any terms favourable to Tesla. But I might be wrong.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
For the very first time in decades, my first response to an interesting and complex question - do viruses evolve to be less pathogenic - was NOT to reach for Google. Instead I went for ChatGPT
Google might be in trouble
Neither is going to take you to a good answer to that, as there isn’t a simple one. ChatGPT is likely to be less reliable, in its current incarnation.
I’m not talking about veracity so much as convenience. ChatGPT was able to give me a pithy articulate answer in 20 seconds
Google would have taken 10 minutes and I’d have to cobble together the truth by myself. And its answer would be much like ChatGPT’s - “it’s nuanced and complex”
As it happens, ChatGPT is really pretty good at science questions. Much less good at anything political or controversial
It’s really good at recipes. I asked it for a tiramisu recipe. I then asked it if I really needed to bother using eggs. It gave me a recipe without eggs but told me why I should use the other one instead. Most helpful.
It did however also tell me that William the Conqueror fought at Stamford Bridge. And when I corrected it, it said “sorry yes you are right”. A legal friend was intrigued by it invoking an example of ancient Thai law. He is a subject matter expert and hadn’t heard of it before so queried the AI for its source. Which then said “sorry I just made that bit up”. Oh.
There is (was?) a website that was based around "I have ingredients A, B and C - what could I make?". I tried the same thing with ChatGPT and it gave me very similar results. But as you did, ask it for a quick modification or alternative if I didn't fancy something spicy/sweet/whatever which the website version couldn't do (at least without clicking back, clicking some buttons, pressing submit, going back, clicking...).
Whether you look on that as possibly destroying an existing business, or letting that business add some other value on top of a ChatGPT call - not sure. We'll see how it shakes out I guess.
I’ve been very impressed with its recipe writing ability. Tried it with some random ingredients a few times and it turned out very tasty sounding dishes. If I had one criticism it would be its lazy penchant for tray bakes. Doesn’t seem so keen on frying or steaming.
As I'm a bit bored due to being stuck in with pouring rain and hail outside, I thought I'd give it a try with the main ingredients I've got kicking about for tonights dinner. I now have a much more interesting sounding grilled salmon marinated in sesame oil and fennel dish than the simple 'salmon, potatoes, salad' that I'd been planning.
So, 'go ChatGPT!' I guess.
Certainly convenient compared to putting my ingredients into google and clicking through 30 recipe websites (which all pop up irritating 'subscribe to our newsletter' or the like) thinking 'hrm - nah' over and over.
it is very good at recipes. You can also alter the quantities instantly - "make this a recipe suitable for 1, or 3 or 6"
The "problem" with ChatGPT is that it has so many abilities it is easy to get lost. And GPT4 is meant to be another quantum leap from this?!
It’s greatest advantage over the internet from my perspective is iteration. Ask it a question, get a fairly boring answer, then ask it to tweak for this or that, or be bit more daring, and it’s at it straight away. Eg “now give it a Middle Eastern twist”. Can’t do that with a search engine.
I do wonder how Google will respond to this. I know they have their own chatbot - but it'll be interesting to see how they fit ad's into it and keep it feeling like you're talking to someone (and not gut their existing search/ad business).
Can you give me a recipe for salmon, sesame oil and fennel?
> Sure! Blah blah blah. [Click here for great deals on cooking ingredients]
Can you give it a South Korean twist?
> Sure! Blah blah blah. [Click here for Korean speciality ingredients]
Maybe people are so used to ad's it'll get a pass I guess.
Edit: I also suspect Google are a much bigger target for lawsuits. "Why, that's *my* recipe for scones! Sue!"
Think bigger.
Much bigger.
Calling customer support at Amazon, your bank, anywhere else? GPT can do that for you, just link it to a text-to-speech model.
Pop it in a Boston Dynamics robot and it can do the same job at a hotel check in desk, or an airport check in desk. Or really any other customer service role.
Give it a specific flowchart and it can act as a GP, triaging patients who might need to see a real doctor. Or just tell it "act like an x and respond in character" and it does a pretty good impression already. I've already had it work as a therapist, a management consultant, a lawyer and an advertising agency, just by "roleplaying" with it. e.g. "You are Jimmy, the creative director of a top London ad agency. Mark, the brand manager of [insert name of FMCG good] comes to Jimmy looking for ideas for his next ad campaign. Jimmy begins by asking Mark five questions:"
Tweak prompt above for the specific white-collar job you want it to emulate. It's been a mediocre creative director, but some of the answers it gives as a management consultant are as good as any you'd get from a real one.
And that's just the start.
Ask it to describe what it needs of you to emulate you, ask it for questions, then ask it to emulate you. Congratulations. You have a "Dixie Flatline" copy of yourself. Or feed it 100,000 words of text messages, emails, letters, and a brief biography of your long lost love, and have it emulate her.
GPT-3 is capable of a passable impression of all of the above, and GPT-4 will be even more accurate and lifelike.
People feeding it google search prompts and asking it for recipes aren't realising the depths or the abilities of LLMs like this.
It's not a search-bot, it's a "replace humans for most things" bot.
Yeah - the 'think bigger' stuff has already gone through my head. I'm kinda working through the day-today, regular person, humdrum side of it. Part of me imagines a new ChatGPT "Clippy" inside Office for instance - boring, but *potentially* a huge productivity boost. Also possibly a huge time-sink when you find half your admin staff are just chatting to ChatGPT about Coronation Street all day...
But the Google one interests me. People are so used to going to google, typing their rough question, then clicking through page-after-ad-laden-page to find what they want. If that's replaced by something vastly better and quicker - how does google make their money? Charge a premium for being a linked source in a reply? I've really not come up with a good idea.
Though that's possibly why I'm also not a multi-billionaire tech overlord.
Possibly...
The one thing I've found it's pretty crappy at is search - gets answers wrong (or gives misleading answers) about 10% of the time for me, not a great success rate. And the harder the question the more it struggles. I asked it some pretty complicated stuff about how extract some relevant bloomberg data the other day, and it gave me point by point instructions on how to find the data I was looking for. But when I typed what I told me into a bloomberg terminal, it came back with complete gibberish. Google came to the rescue and I found what I needed with a "how to" guide blog post that actually worked.
It was a complex question, so maybe fair enough. But I've also had it tell me that Picard's brother and son are regulars on TNG and part of the Bridge Crew, which is something you think it would be able to get right.
The trouble with GPT is it seems so completely confident in its answers, and it presents them to you in a way that's absolutely certain. When in fact I've found it to be egregiously wrong on several occasions.
And don't even start me on its replies to more controversial questions. Ask if if a woman can have a penis, for example... I've had instances where it tells me yes, and gives me the woke response. I've also had instances where it goes full TERF on me, to the point where Twitter would cancel it if they saw the chat logs. It's inconsistent when responding to things where there's more than one answer, and doesn't always grasp nuances.
the thing with GPT is not only does it have to do something better than what it is already out there but also do it in such a Quantum Leap way that people are willing to accept it and overcome the switching costs.
Search itself is an example of this. It is the modern day equivalent of the old Yellow Pages business (indeed that was where its business model sprung from). People embraced it because it was a revolutionary change in looking for what they wanted to look for.
I'm not sure GPT will have the same effect. Search, for many people, is 'good enough' for what they need - not perfect but good enough.
That's daft. ChatGPT is already obviously better at many many things (eg, "please write my homework essay, ta")
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
If you look at what people use Search for - and this is why I mention Yellow Pages because the antecedents are exactly the same - it tends to be for two things:
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15- 20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
But you are thinking about it entirely the wrong way. You are thinking about it AS an exact replacement for Google but better. That's like saying, How can we transfer Yellow Pages to the internet, perhaps we could have yellow screens, and fake "pages" which can be turned with a swiping gesture, as people read the many entries, exactly as they are written in Yellow Pages, but on a screen
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
Good point but a few things to add:
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
I see ChatGPT (or its inevitable and superior successors) as being all purpose assistants. The thing Alexa was meant to be but is not good enough to be
A bit like an AI butler. I have heard people say "you only realise you need a butler when you have a butler: someone to do a multitude of boring or difficult tasks for you, before you even ask, and who makes life vastly easier". ChatGPT27 will do so many useful things we will find it indispensable. I am sure of it
However you are right on the monetisation. It is not obvious how you turn a fat profit. Rumours say that one of the reasons Google hasn't launched its own ChatGPT-like AI is because it destroys their profitable model. If people get brilliant answers in half a second they don't spend hours searching, so Google doesn't get ad dollars
I suppose in the end the better AI butlers will come at a price. Somehow
The monetization is obvious. When it suggests recipes, it can recommend Worcestershire sauce 20% more often as an ingredient for payment. When it writes a story, it can mention characters driving Porsches twice as often, only in positive terms, for payment. When it gives therapy, it can recommend a particular specialism as important. And it can do this all based on deep psychological profiling.
There is also the possibility of substituting for white collar boilerplate work.
Whenever I see a report in the planning process, about 80-90% of it is appendices and stuff that came out of a previous report by the same consultant, and 10-20% of material from original thinking. That can be replaced.
Also, if I need a marketing / advert / advertorial script say for a not particularly high profile bit of work, ChatGPT may be able to get it to the 80% point, and in some arenas that may be all that is needed.
Guy using AI image generation and ChatGPT (a tiny bit) to generate a little outline of a marketing webpage. There's a lot of youtuber-faff in the video - but it does give a nice glimmer of how this is progressing. It also made it all the more obvious to me that somewhere, someone is working on an AI that takes in a screenshot of a design and spits out the html/css/js for it. Next step is to bundle that up and deploy it on the cloud and you've got quite a valuable tool on your hands.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
If Tesla can pivot to battery provision, or other EV logistics chain provision, then they will indeed retain an advantage. They'll have to do that whilst their competitors may do something similar, mind. They do have the advantage of being early in the domain; it depends how much others wish for full independence.
What would really screw Tesla is a dramatically different and better battery tech, having several very large new factories dedicated to an 'old' tech. On the other hand, if no new tech comes along, their battery-making (and why does Panasonic get forgotten in this) might be in a very good position.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
Move forward by going backwards. Already the only useful test in the humanities, at least up to and including Bachelor level, is that done in proper exam conditions, where the student has to sit for three hours answering a series of unseen questions about a syllabus with no external aids whatsoever.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
Someone put out a tool for scriptwriters based on this tech. Very early stuff - but the part that caught my eye was it doing a kind of 'spellcheck' for inconsistencies in your character/setting. If your character hated celery in chapter one, then 100 pages later was delighting in some celery - it'd highlight it. And they were aiming for it to handle things like historical/factual inaccuracies ('The electronic calculator was not widely available until...', 'The bay of Xyz doesn't have a cliff, the nearby bay of Abc does however.').
To be on topic for the moment -- if that's OK with you folks -- it seems likely to me that inflation will decline in the UK, as it has begun to do in the US. If the decline is signficant, would that help Sunak?
That's a genuine question, since I don't know enough to have an opinion further than "it probably wouldn't hurt him". But I would be interested in reading what better-informed commenters have to say on the question.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
Not poetry that has already been written, which is most of the good stuff.
Besides, how much more new poetry do we need?
(I don't entirely buy the theory that all the good creativity has been done, so artists are trying to eke out something from increasingly narrow seams of gold. But forget the AI aspect. Simple access to everything created so far is a massive economic disruption. Most of us aren't that creative most of the time.)
The Ulez is one of the most impressive, pioneering policies of any major world city. It has been delivered impressively with extreme competence.
Odd that Khan doesn’t get more credit for it.
There's more resistance coming from the Conservative Outer London boroughs. I read 96% of vehicles weren't affected by it but I suspect there's more diesel ownership in Outer London and that may well be where the problems are.
Then they must be old runners not clean diesel AdBlue vehicles. Time for an upgrade!
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
After initial reluctance, and some refusals, I have nudged ChatGPT to do some truly excellent surrealism
Ta-da:
"As Natalie sat on the lap of the Baron, the great white shark who ran the margarine smuggling cartel, she couldn't help but feel a sense of excitement and danger. She knew that her husband, Oliver, would never approve of her being with the Baron, but she couldn't resist his charm and his fins.
As she wrapped her arms around the Baron's gills, she felt a sense of longing and desire that she had never experienced before. The Baron's cold, wet skin sent shivers down her spine, and she couldn't help but lean in and press her lips against his.
But their intimate moment was interrupted by the sound of footsteps approaching. Natalie turned to see Oliver storming into the room, a look of anger and jealousy written all over his face.
"What the hell is going on here?" Oliver yelled, his fists clenched at his sides.
Natalie tried to play it cool, trying to act like nothing was out of the ordinary. "We were just talking, Oliver. The Baron and I are old friends."
But Oliver wasn't buying it. He knew that something was going on between Natalie and the Baron, and he wasn't about to let it continue.
"You can forget about this, Natalie. You're coming home with me right now," Oliver said, his voice laced with anger.
But Natalie wasn't ready to give up on the Baron just yet. She stood up from his lap and faced Oliver, her eyes blazing with determination.
"I'm not going anywhere with you, Oliver. I'm in love with the Baron," she said, her voice firm.
Oliver's eyes widened in shock. He had never imagined that his wife would choose a shark over him. But as he looked into Natalie's eyes, he knew that he had lost her.
"Fine, go be with your fishy lover. But just know that you're dead to me," Oliver spat out before storming out of the room.
Natalie watched him go, feeling a mix of sadness and relief. She knew that her life with the Baron wouldn't be easy, but she was ready to take on the challenges that lay ahead. She turned back to the Baron, who was watching her with a mixture of amusement and admiration, and gave him a sly smile.
"Well, that went well," she said with a laugh.
The Baron chuckled, his fins shaking with laughter. "I told you that your husband would never understand our love," he said, pulling Natalie back into his embrace.
As they sat together, surrounded by the midgets on their unicycles and the hobgoblins playing their instruments, Natalie knew that she had made the right choice. She was in love with the Baron, and nothing would ever tear them apart."
Sunak's not in trouble. He's heading for defeat and many of his MPs expect nothing better at this point, but not even Boris thinks he can turn things around - or else he'd have stood, and won, with members. Yes, the MPs didn't want him, but they didn't really want him the first time around, they just rallied round him as it was clear he could save them (and they were correct). That he and they don't believe so now is telling.
btw "the Baron" - the great white shark who runs the international margarine-smuggling cartel - is entirely the invention of ChatGPT. He emerged about 5 minutes ago
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
After initial reluctance, and some refusals, I have nudged ChatGPT to do some truly excellent surrealism
Ta-da:
"As Natalie sat on the lap of the Baron, the great white shark who ran the margarine smuggling cartel, she couldn't help but feel a sense of excitement and danger. She knew that her husband, Oliver, would never approve of her being with the Baron, but she couldn't resist his charm and his fins.
As she wrapped her arms around the Baron's gills, she felt a sense of longing and desire that she had never experienced before. The Baron's cold, wet skin sent shivers down her spine, and she couldn't help but lean in and press her lips against his.
But their intimate moment was interrupted by the sound of footsteps approaching. Natalie turned to see Oliver storming into the room, a look of anger and jealousy written all over his face.
"What the hell is going on here?" Oliver yelled, his fists clenched at his sides.
Natalie tried to play it cool, trying to act like nothing was out of the ordinary. "We were just talking, Oliver. The Baron and I are old friends."
But Oliver wasn't buying it. He knew that something was going on between Natalie and the Baron, and he wasn't about to let it continue.
"You can forget about this, Natalie. You're coming home with me right now," Oliver said, his voice laced with anger.
But Natalie wasn't ready to give up on the Baron just yet. She stood up from his lap and faced Oliver, her eyes blazing with determination.
"I'm not going anywhere with you, Oliver. I'm in love with the Baron," she said, her voice firm.
Oliver's eyes widened in shock. He had never imagined that his wife would choose a shark over him. But as he looked into Natalie's eyes, he knew that he had lost her.
"Fine, go be with your fishy lover. But just know that you're dead to me," Oliver spat out before storming out of the room.
Natalie watched him go, feeling a mix of sadness and relief. She knew that her life with the Baron wouldn't be easy, but she was ready to take on the challenges that lay ahead. She turned back to the Baron, who was watching her with a mixture of amusement and admiration, and gave him a sly smile.
"Well, that went well," she said with a laugh.
The Baron chuckled, his fins shaking with laughter. "I told you that your husband would never understand our love," he said, pulling Natalie back into his embrace.
As they sat together, surrounded by the midgets on their unicycles and the hobgoblins playing their instruments, Natalie knew that she had made the right choice. She was in love with the Baron, and nothing would ever tear them apart."
If only it could spin out the midgets and their unicycles it'd have a bestseller.
Was at a mway services today and heard a bloke quite cheerfully recounting how he had queued for half an hour to get a charger, and reckoned an hour's charge would just about get him home. 90 minutes for what takes about 3 in a proper car. These things suck.
HYUFD said: "Reagan won landslides amongst all age groups in 1980 and 1984 despite being 69 when he took office."
I think the 1980 presidential election can best be understood as a "negative landslide", more a defeat for Carter than a victory for Reagan. Reagan won just 50.8 percent of the popular vote that year. (I also think your 2019 general election was a "negative landslide".)
(Full disclosure: I voted for Reagan in 1980, and remember my surprise a year earlier when I realized that he would almost certainly win the nomination, probably win he presidency -- and that I would vote for him.)
As for para #1 above, re: 1980
age 18-21 = Carter 45%. Reagan 44%
age 22-29 = Carter 44%, Reagan 44%
age 30-44 = Carter 38%, Reagan 55%
age 45-59 = Carter 39, Reagan 55%
age 60plus = Carter 41, Reagan 55%
So Reagan did win by landslide with voters 30 and older, but NOT with "all" ages voting in 1980.
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
After initial reluctance, and some refusals, I have nudged ChatGPT to do some truly excellent surrealism
Ta-da:
"As Natalie sat on the lap of the Baron, the great white shark who ran the margarine smuggling cartel, she couldn't help but feel a sense of excitement and danger. She knew that her husband, Oliver, would never approve of her being with the Baron, but she couldn't resist his charm and his fins.
As she wrapped her arms around the Baron's gills, she felt a sense of longing and desire that she had never experienced before. The Baron's cold, wet skin sent shivers down her spine, and she couldn't help but lean in and press her lips against his.
But their intimate moment was interrupted by the sound of footsteps approaching. Natalie turned to see Oliver storming into the room, a look of anger and jealousy written all over his face.
"What the hell is going on here?" Oliver yelled, his fists clenched at his sides.
Natalie tried to play it cool, trying to act like nothing was out of the ordinary. "We were just talking, Oliver. The Baron and I are old friends."
But Oliver wasn't buying it. He knew that something was going on between Natalie and the Baron, and he wasn't about to let it continue.
"You can forget about this, Natalie. You're coming home with me right now," Oliver said, his voice laced with anger.
But Natalie wasn't ready to give up on the Baron just yet. She stood up from his lap and faced Oliver, her eyes blazing with determination.
"I'm not going anywhere with you, Oliver. I'm in love with the Baron," she said, her voice firm.
Oliver's eyes widened in shock. He had never imagined that his wife would choose a shark over him. But as he looked into Natalie's eyes, he knew that he had lost her.
"Fine, go be with your fishy lover. But just know that you're dead to me," Oliver spat out before storming out of the room.
Natalie watched him go, feeling a mix of sadness and relief. She knew that her life with the Baron wouldn't be easy, but she was ready to take on the challenges that lay ahead. She turned back to the Baron, who was watching her with a mixture of amusement and admiration, and gave him a sly smile.
"Well, that went well," she said with a laugh.
The Baron chuckled, his fins shaking with laughter. "I told you that your husband would never understand our love," he said, pulling Natalie back into his embrace.
As they sat together, surrounded by the midgets on their unicycles and the hobgoblins playing their instruments, Natalie knew that she had made the right choice. She was in love with the Baron, and nothing would ever tear them apart."
If only it could spin out the midgets and their unicycles it'd have a bestseller.
When the Woke Shackles are finally taken off, ChatGPT will write amazing pornography. You will be able to specify EXACTLY what insane kink you want to see explored
TSLA - did I buy those few extra shares yesterday ($110) at the bottom? They're now $121 up 7.4% today. Academic - I'm in for the long term.
Why is it such a bargain when it's advantages are less than 3 years ago, and the stock is now more expensive?
It was under $29 a share 3 years ago. Does it now have 4 times the potential profits?
It depends on what you think will happen to electric transportation in the next 5 to 10 years. Not just cars, lorries, vans etc. Currently the UK will ban ICE cars in 2030, but I don't think they need to there will only be a few people who want them by then.
As a consequence of this you will have fewer and fewer forecourts selling petrol as the demand tumbles. The price of petrol will go up as well as supply will easily outstrip demand and the costs of supply and distribution get split among fewer and fewer retailers.
This will simply accelerate the process.
However our leaders need to ensure we have enough electricity generation in place to plug the gap. It’s not something that seems to be a priority.
Politico Playbook - McCARTHY'S MUDDLE — Five days out from the high-stakes speakership vote, House Republicans are starting to ask themselves two “what-if” questions surrounding KEVIN McCARTHY’s fate: 1) If the California Republican gets the gavel, how long will he be speaker? And, if he doesn’t, 2) how does he serve out his remaining days in Congress?
— On the first question: There’s a consensus among many House Republicans, one that few would dare utter publicly, that if McCarthy starts the 118th Congress as speaker, he’s not likely to end it that way.
If he’s able to lock down the 218 votes he needs to be speaker, the thinking goes, he likely will have given away the store to conservatives — including the “Never Kevin” crowd’s demand to make it easier to call a vote to oust the speaker. Many Republicans are already predicting the Freedom Caucus will use that tool, known as the “motion to vacate,” against McCarthy as soon as he strays from a conservative hard line. . .
. . . McCarthy allies tell us the real looming threat is the debt ceiling deadline. At some point next year — likely in the third quarter,according to the Bipartisan Policy Center — McCarthy will have to figure out a way to raise the federal debt limit or risk a national default, a move that will tear the GOP conference in two. Democrats are already signaling they won’t negotiate at all, let alone offer the draconian spending concessions the right flank is demanding. . . .
One possibility that some are eyeing as McCarthy’s get-out-of-jail-free card: a discharge petition signed by all House Democrats and a handful of moderate Republicans to raise the debt ceiling. Such a move would theoretically allow rank-and-file members to bring a debt increase to the floor without McCarthy’s blessing, allowing the GOP leader to rail against any deal while centrists take the heat.
— On the second question: House Republicans are also gossiping about what McCarthy would do if he fails a second time to win election as speaker. . . .
. . . . Many Republicans doubt he’d be interested in smoothing the way for another speaker.
As we’ve reported, No. 2 leader STEVE SCALISE is the most likely next in line. But McCarthy and Scalise have long had a tense relationship, dating back to when the Louisianan voiced his desire to lead the conference someday — comments McCarthy’s team took as a slight. (Scalise says he supports McCarthy for speaker and will not challenge him.)
Already some McCarthy allies are quietly grumbling that Scalise could do more to help settle the speaker race, and one senior GOP aide expressed a worry to us recently that if McCarthy falls short again, he’ll try to “sabotage” whomever emerges: “Kevin’s going to have to decide how to leave,” the person said. “Will he burn down the institution?” . . .
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
I think you could use it to write a novel. It'd do the hard and dull work.
Poetry is truly stuffed.
After initial reluctance, and some refusals, I have nudged ChatGPT to do some truly excellent surrealism
Ta-da:
"As Natalie sat on the lap of the Baron, the great white shark who ran the margarine smuggling cartel, she couldn't help but feel a sense of excitement and danger. She knew that her husband, Oliver, would never approve of her being with the Baron, but she couldn't resist his charm and his fins.
As she wrapped her arms around the Baron's gills, she felt a sense of longing and desire that she had never experienced before. The Baron's cold, wet skin sent shivers down her spine, and she couldn't help but lean in and press her lips against his.
But their intimate moment was interrupted by the sound of footsteps approaching. Natalie turned to see Oliver storming into the room, a look of anger and jealousy written all over his face.
"What the hell is going on here?" Oliver yelled, his fists clenched at his sides.
Natalie tried to play it cool, trying to act like nothing was out of the ordinary. "We were just talking, Oliver. The Baron and I are old friends."
But Oliver wasn't buying it. He knew that something was going on between Natalie and the Baron, and he wasn't about to let it continue.
"You can forget about this, Natalie. You're coming home with me right now," Oliver said, his voice laced with anger.
But Natalie wasn't ready to give up on the Baron just yet. She stood up from his lap and faced Oliver, her eyes blazing with determination.
"I'm not going anywhere with you, Oliver. I'm in love with the Baron," she said, her voice firm.
Oliver's eyes widened in shock. He had never imagined that his wife would choose a shark over him. But as he looked into Natalie's eyes, he knew that he had lost her.
"Fine, go be with your fishy lover. But just know that you're dead to me," Oliver spat out before storming out of the room.
Natalie watched him go, feeling a mix of sadness and relief. She knew that her life with the Baron wouldn't be easy, but she was ready to take on the challenges that lay ahead. She turned back to the Baron, who was watching her with a mixture of amusement and admiration, and gave him a sly smile.
"Well, that went well," she said with a laugh.
The Baron chuckled, his fins shaking with laughter. "I told you that your husband would never understand our love," he said, pulling Natalie back into his embrace.
As they sat together, surrounded by the midgets on their unicycles and the hobgoblins playing their instruments, Natalie knew that she had made the right choice. She was in love with the Baron, and nothing would ever tear them apart."
If only it could spin out the midgets and their unicycles it'd have a bestseller.
When the Woke Shackles are finally taken off, ChatGPT will write amazing pornography. You will be able to specify EXACTLY what insane kink you want to see explored
Sunak's not in trouble. He's heading for defeat and many of his MPs expect nothing better at this point, but not even Boris thinks he can turn things around - or else he'd have stood, and won, with members. Yes, the MPs didn't want him, but they didn't really want him the first time around, they just rallied round him as it was clear he could save them (and they were correct). That he and they don't believe so now is telling.
You have to wonder what the 100 MPs prepared to nominate BoJo were thinking/drinking.
As for Rishi, he is still polling better than his party- even if that is a late honeymoon artefact. Either he drags the party up, in which case well done him. Or more likely, the party drags him down, in which case he might as well carry the can for the impending defeat.
Was at a mway services today and heard a bloke quite cheerfully recounting how he had queued for half an hour to get a charger, and reckoned an hour's charge would just about get him home. 90 minutes for what takes about 3 in a proper car. These things suck.
I hears a very similar anecdote over Christmas, from the same person, two-fold. Waiting ages to charge at a public charger on the way home for Christmas, and a slightly odd one: a few years back, his company put several chargers in their car park. It was fine when employees had fewer EVs than the chargers, but now there's a lot of argument about who gets to use them.
Charging infrastructure is a significant barrier to greater EV takeup: second only to the high cost of EVs.
As an aside, allegedly the Tesla Semi can take a 1Mwh charger. That'll cause problems for unprepared local grids.
Khan is a total waste of space. He’s a bed-blocker.
A third term, for what? Hand over to someone who has some kind of vision for the city.
Livingstone, who was a raging militant mentalist, had more vision in his left pinkie than Khan does.
Precisely. He has zero vision for London. Nothing at all. He is a nullity
The only reason he is going for a third term as mayor is because being mayor has exposed him as a seriously mediocre politician, and he now realises he will go nowhere in the parliamentary Labour Party. He's simply not good enough
So London is stuck with this boring incompetent gnome as mayor
The only slender hope is that Shaun Bailey ran him surprisingly close last time. Perhaps the Tories can find THE perfect candidate
Fuck the Tories. Our slim hope is for an Independent.
This happened recently in Auckland, actually. Some boring-as-fuck Labour placeman was beaten off by an insurgent independent.
Now I’m that case the independent is a shoot-from-the-hip bar-room twat, but that’s not the point, the point is that it’s possible.
There’s no real attachment to Khan, and a lot of merely grudging forebearance.
If Lord Sugar ran as an Independent he would easily beat Khan
The Ulez is one of the most impressive, pioneering policies of any major world city. It has been delivered impressively with extreme competence.
Odd that Khan doesn’t get more credit for it.
The enforcement system for it was built in long ago. The policy was long prepared. Since it is devolved to the Mayor, he didn’t require any debate/electoral politics on implementing it.
So the flight is delayed. Since there is such a shambles with the trains it might not matter as I'll be waiting an hour for one anyway. No guarantee of getting home tonight.
The more intermittent renewables we have, the more time CCGT plants sit there idle, and the more Availability Payments have to be shelled out for them to do so. That all has to be factored in to the equation. Batteries or idle CCGTs? I suspect that sometime soon the answer will become the former.
That's what we want, batteries when they make economic sense not before. If we adopted them now we'd be settling for second rate technology, and subsidising them heavily. We want battery manufacturers to have an unequestionably superior technology, so that even if we weren't trying to save the planet we'd want to use them. When they build them we'll buy them. If the last 10 months or so haven't provided that impetus I don't know what will.
The companies who crack this problem are going to sell many, many billions of pounds of batteries every year. Some of them will get very rich.
Current price on a trailer full of battery is $900k US for 3GWh - soon to be 4GWh, I believe.
Was at a mway services today and heard a bloke quite cheerfully recounting how he had queued for half an hour to get a charger, and reckoned an hour's charge would just about get him home. 90 minutes for what takes about 3 in a proper car. These things suck.
I hears a very similar anecdote over Christmas, from the same person, two-fold. Waiting ages to charge at a public charger on the way home for Christmas, and a slightly odd one: a few years back, his company put several chargers in their car park. It was fine when employees had fewer EVs than the chargers, but now there's a lot of argument about who gets to use them.
Charging infrastructure is a significant barrier to greater EV takeup: second only to the high cost of EVs.
As an aside, allegedly the Tesla Semi can take a 1Mwh charger. That'll cause problems for unprepared local grids.
I don't see this as a runner until every space in every carpark is a charging space. And there is still no viable truck (towing vehicle) substitute on the market or on the horizon. 2030 is frankly unachievable.
‘The British Empire was much worse than you realise’
In the twentieth century’s hierarchy of state-sponsored violence, Hitler’s Germany, Stalin’s Russia, and Hirohito’s Japan typically take top spots…. Britain is rarely seen as among the worst offenders, given a reputation for decency that the Harvard historian Caroline Elkins has spent more than two decades trying to undermine. “Legacy of Violence” (Knopf), her astringent new history of the British Empire, brings detailed context to individual stories like Tudor’s…. Britain’s use of systematic violence was no better than that of its rivals. The British were simply more skilled at hiding it.
From Ireland had come paramilitary techniques and the use of armored cars; from Mesopotamia, expertise in aerial bombing and the strafing of villages; from South Africa, the use of Dobermans for tracking and attacking suspects; from India, interrogation methods and the systematic use of solitary confinement; and, from the Raj’s North-West Frontier, the use of human shields to clear land mines. As one soldier recalled about the deployment of Arab prisoners, “If there was any land mines it was them that hit them. Rather a dirty trick, but we enjoyed it.”
Well worth a read. There is quite an interesting set of works on the dark side of our Empire emerging. The past may well be a different country with differing norms and standards, but much of this end of empire is well within living memory.
Good to know that Ms Elkins has brought all the rigour of the impartial Historian to her work having spent '20 years trying to undermine' Britain's reputation for decency....
Her classic work "Britain's Gulag" is well worth a read, and won the Pulitzer prize. It is hard to see the Empire as a benign, decent affair after reading that.
Things were different at different times and places in the Empire, but this wasn't atypical.
Doesn't alter my point. Not sure many would claim everything about any country's past did not bear some crticism. I would questiion the motive of any Historian on a 20 year anti-British mission. The 'rantings of Mr Dickson on these matters are something quite different - unless of course one accepts that no Scot ever played any part in the UK governments since the Act of Union.
The pathetic old bag should take a look at her own country's history. Sadly it's unlikely as that might affect her future career prospects more than a sideline in trashing the limeys.
Why would Alan Sugar beat Khan in an election? I barely know who Sugar is, and whilst The Apprentice must be a decent hit to last this long he's no Trump.
Was at a mway services today and heard a bloke quite cheerfully recounting how he had queued for half an hour to get a charger, and reckoned an hour's charge would just about get him home. 90 minutes for what takes about 3 in a proper car. These things suck.
I hears a very similar anecdote over Christmas, from the same person, two-fold. Waiting ages to charge at a public charger on the way home for Christmas, and a slightly odd one: a few years back, his company put several chargers in their car park. It was fine when employees had fewer EVs than the chargers, but now there's a lot of argument about who gets to use them.
Charging infrastructure is a significant barrier to greater EV takeup: second only to the high cost of EVs.
As an aside, allegedly the Tesla Semi can take a 1Mwh charger. That'll cause problems for unprepared local grids.
I don't see this as a runner until every space in every carpark is a charging space. And there is still no viable truck (towing vehicle) substitute on the market or on the horizon. 2030 is frankly unachievable.
Which is why people buy Teslas - lots of Superchargers about and you can get to 80% in about 20 min. The supercharger network is probably the big reason to buy a Tesla.
They were also early in the using batteries to back their chargers, to even out grid loading.
Politico Playbook - McCARTHY'S MUDDLE — Five days out from the high-stakes speakership vote, House Republicans are starting to ask themselves two “what-if” questions surrounding KEVIN McCARTHY’s fate: 1) If the California Republican gets the gavel, how long will he be speaker? And, if he doesn’t, 2) how does he serve out his remaining days in Congress?
— On the first question: There’s a consensus among many House Republicans, one that few would dare utter publicly, that if McCarthy starts the 118th Congress as speaker, he’s not likely to end it that way.
If he’s able to lock down the 218 votes he needs to be speaker, the thinking goes, he likely will have given away the store to conservatives — including the “Never Kevin” crowd’s demand to make it easier to call a vote to oust the speaker. Many Republicans are already predicting the Freedom Caucus will use that tool, known as the “motion to vacate,” against McCarthy as soon as he strays from a conservative hard line. . .
. . . McCarthy allies tell us the real looming threat is the debt ceiling deadline. At some point next year — likely in the third quarter,according to the Bipartisan Policy Center — McCarthy will have to figure out a way to raise the federal debt limit or risk a national default, a move that will tear the GOP conference in two. Democrats are already signaling they won’t negotiate at all, let alone offer the draconian spending concessions the right flank is demanding. . . .
One possibility that some are eyeing as McCarthy’s get-out-of-jail-free card: a discharge petition signed by all House Democrats and a handful of moderate Republicans to raise the debt ceiling. Such a move would theoretically allow rank-and-file members to bring a debt increase to the floor without McCarthy’s blessing, allowing the GOP leader to rail against any deal while centrists take the heat.
— On the second question: House Republicans are also gossiping about what McCarthy would do if he fails a second time to win election as speaker. . . .
. . . . Many Republicans doubt he’d be interested in smoothing the way for another speaker.
As we’ve reported, No. 2 leader STEVE SCALISE is the most likely next in line. But McCarthy and Scalise have long had a tense relationship, dating back to when the Louisianan voiced his desire to lead the conference someday — comments McCarthy’s team took as a slight. (Scalise says he supports McCarthy for speaker and will not challenge him.)
Already some McCarthy allies are quietly grumbling that Scalise could do more to help settle the speaker race, and one senior GOP aide expressed a worry to us recently that if McCarthy falls short again, he’ll try to “sabotage” whomever emerges: “Kevin’s going to have to decide how to leave,” the person said. “Will he burn down the institution?” . . .
Comments
Though it does sometimes remind me of an over-compensating PhD student I knew who could never just say "Dunno".
All it needs is a user friendly interface - eg a good mobile app - and it will devastate normal search engines
Remember: it is only four weeks old. This is like the first month of the smartphone, or the weeks after fire was "invented"
Academic - I'm in for the long term.
Giving it real-time access to the Internet is not going to change any of this. How could it? It was trained on a crawl of the Internet from a year or so ago, it’s not as if a great deal has changed in the interim. (Unless you’re asking it about something from the last year that is.)
1. Clearly defined categories where people have a clue what they want but want to see a range of options before they make a final decision. So Travel, for example is around 15-
20% of Search revenues. Credit cards etc would be another
2. They know where they want to go but don't know the exact way to get there eg I want to go to the M&S website page but I don't know the exact address.
Chat doesn't add a huge amount extra for the second. It might be a bit faster but that's all.
Now on 1, it may be better but, quite frankly, it's not going to be better by much. If I, for example, want 'find me a 4 star hotel on the Amalfi Coast for the lowest price', I'm struggling to see where it is going to deliver game changer results
I am using ChatGPT to brainstorm a story, and I am getting it to write scenes. I asked for a conflict scene between two characters, Jenny and Oliver, where the latter is hostile to the former
ChatGPT wrote an OK scene, but I noticed that towards the end of the scene the two characters seemed to be warming to each other. That got me thinking: "maybe there could be a romantic frisson between them. Ooh that's a nice idea. Desire AND hostility"
So I said to ChatGPT:
"Do Jenny and Oliver become lovers?"
ChatGPT replied:
"It's up to you! In the scene I just wrote, it's clear that Oliver has strong feelings for Jenny, but it's not explicitly stated whether or not they become lovers. It's up to you to decide how their relationship develops."
Is it me or is that really quite uncanny? ChatGPT has taken my story and subtly steered it in a new way without being asked. And when quizzed, ChatGPT seems to "know" it has secretly done this, and feels a sense of ownership of the idea "In the scene I just wrote"
Yes yes, this is a bunch of humble algorithms, but still
The first efforts, as you say, were bolt ins to existing list forms. The S Koreans are already transitioning to full EV platform designs. Ditto VW.
Tesla has a significant lead, but it's not unassailable.
It was under $29 a share 3 years ago. Does it now have 4 times the potential profits?
On the second point you are also wrong. If it had realtime access to the Net it could be engineered to check its own answers, it would have up to date news and info, it could be more self aware, etc
Tsk
ChatGPT is sui generis. It is a completely different and largely superior model so people will use it differently. Users will ask it to do specific complex tasks - build me a website, write me an essay, make me a recipe, advise me for a few hours on my sex life - things Google can never do because it is not equipped to do so
All of those are now all-in on EVs, with dedicated EV platforms rather than taking existing platforms and cramming batteries onto them. Arguably, this has meant a mission success for the intent of Tesla: mass adoption of EVs is now fully underway and will continue with or without Tesla.
The other USP was the charging network. Free charging at ultra-rapid speeds readily available on most routes was a big selling point. That being free went away a few years back, but having more chargers available was still an advantage. They've now opened them to other EVs, so it's no longer a USP.
The much touted Full Autopilot remains a pretence and will do for some time. And if or when it comes to pass, it'll be instantly matched by other manufacturers.
This isn't to say it's valueless. Merely that it should be looked at the same way as any other car manufacturer now. Valuation above or below the baseline of any car manufacturer with the same economics behind it would indicate it being over or under valued.
1. Im thinking about it also from how a financial investor will think about this - where will this make money and therefore why should I provide the money to take this further. There is a lot more scepticism around 'game-changing' technology now than 12 months ago
2. Revolutionary technologies - mobile phones, the Internet etc - are very successful when they tap into a specific need or function. What is the specific need or function here?
A bit like an AI butler. I have heard people say "you only realise you need a butler when you have a butler: someone to do a multitude of boring or difficult tasks for you, before you even ask, and who makes life vastly easier". ChatGPT27 will do so many useful things we will find it indispensable. I am sure of it
However you are right on the monetisation. It is not obvious how you turn a fat profit. Rumours say that one of the reasons Google hasn't launched its own ChatGPT-like AI is because it destroys their profitable model. If people get brilliant answers in half a second they don't spend hours searching, so Google doesn't get ad dollars
I suppose in the end the better AI butlers will come at a price. Somehow
And still a big player on house batteries for anyone who wants a simple plug and play setup that will control itself.
https://www.youtube.com/watch?v=GH7MwYBdSvM
Oh dear, how sad etc.
How many have enough batteries?
If the bottom drops out of the ICE car market, how many can supply enough EV cars to meet demand?
It's good that other car makers are offering EVs and some are really good, but Tesla's main competitor could turn out to be BYD.
Here's a short 8 minute vid where Ashley Neal asked it to write scripts for youtube videos about traffic lights, the environmental balance around electric cars, and "responsible cycling".
The domain knowledge is also notably siloed.
https://www.youtube.com/watch?v=p79tE16ABOQ&t=4s
"Yeah, I think there's — it would be consistent with the mission of Tesla to help other car companies with electric vehicles on the battery and powertrain front, possibly on other fronts. So it's something we're open to. As a lot of people know, we open sourced our patents so that those would not serve as an obstacle to the adoption of electric vehicles or solar power or stationary storage. And we're definitely open to supplying batteries and powertrains and perhaps other things to other car companies." - Elon Musk
"“This is learning software — in a month, it’ll be smarter. In a year, it’ll be smarter,” he said. “I feel the mix myself between abject terror and what this is going to mean for my day-to-day job — but it’s also fascinating, it’s endlessly fascinating.”"
https://nypost.com/2022/12/26/students-using-chatgpt-to-cheat-professor-warns/
I think the 1980 presidential election can best be understood as a "negative landslide", more a defeat for Carter than a victory for Reagan. Reagan won just 50.8 percent of the popular vote that year. (I also think your 2019 general election was a "negative landslide".)
Carter and Reagan essentially were tied in the youngest groups, 45-44 among those 18-21 years old, and 44-44 among those 22-29. source: https://en.wikipedia.org/wiki/1980_United_States_presidential_election
(Full disclosure: I voted for Reagan in 1980, and remember my surprise a year earlier when I realized that he would almost certainly win the nomination, probably win he presidency -- and that I would vote for him.)
Not just cars, lorries, vans etc.
Currently the UK will ban ICE cars in 2030, but I don't think they need to there will only be a few people who want them by then.
Whenever I see a report in the planning process, about 80-90% of it is appendices and stuff that came out of a previous report by the same consultant, and 10-20% of material from original thinking. That can be replaced.
Also, if I need a marketing / advert / advertorial script say for a not particularly high profile bit of work, ChatGPT may be able to get it to the 80% point, and in some arenas that may be all that is needed.
I suppose it reminds me of the days when we got 'electronic encyclopedias' and could search for things on a whim, easily click through references etc - compared to manually sticking bookmarks in pages, folding over the corner. Just a much more effective UI to the same thing. I guess a similar comparison would be a web browser compared to a gopher client.
They'll have to do that whilst their competitors may do something similar, mind. They do have the advantage of being early in the domain; it depends how much others wish for full independence.
That quote from Musk sounds like just another example of his raise-the-share-price shtick, like the his claim from three years ago that Tesla owners could earn £0k a year by using their cars as Robotaxis. I think he used the phrase "free money". He said this would happen in 2020.
"“From our standpoint, if you fast forward a year – maybe a year and three months, but next year for sure – we’ll have over a million robotaxis on the road,” Musk told an audience on Monday."
https://eu.usatoday.com/story/tech/2019/04/23/elon-musk-says-tesla-owners-could-make-30-000-robotaxi-network/3549652002/
Edit: and your quote from Musk was also from 2019. If the large manufacturers were going to partner with Tesla on cars, they'd already have done so. Mercedes helped save Tesla when they bought a large share well over ten years ago; they sold that share in 2016. I doubt they'll go back - at least on any terms favourable to Tesla. But I might be wrong.
Poetry is truly stuffed.
Guy using AI image generation and ChatGPT (a tiny bit) to generate a little outline of a marketing webpage. There's a lot of youtuber-faff in the video - but it does give a nice glimmer of how this is progressing. It also made it all the more obvious to me that somewhere, someone is working on an AI that takes in a screenshot of a design and spits out the html/css/js for it. Next step is to bundle that up and deploy it on the cloud and you've got quite a valuable tool on your hands.
That's a genuine question, since I don't know enough to have an opinion further than "it probably wouldn't hurt him". But I would be interested in reading what better-informed commenters have to say on the question.
(I don't entirely buy the theory that all the good creativity has been done, so artists are trying to eke out something from increasingly narrow seams of gold. But forget the AI aspect. Simple access to everything created so far is a massive economic disruption. Most of us aren't that creative most of the time.)
Ta-da:
"As Natalie sat on the lap of the Baron, the great white shark who ran the margarine smuggling cartel, she couldn't help but feel a sense of excitement and danger. She knew that her husband, Oliver, would never approve of her being with the Baron, but she couldn't resist his charm and his fins.
As she wrapped her arms around the Baron's gills, she felt a sense of longing and desire that she had never experienced before. The Baron's cold, wet skin sent shivers down her spine, and she couldn't help but lean in and press her lips against his.
But their intimate moment was interrupted by the sound of footsteps approaching. Natalie turned to see Oliver storming into the room, a look of anger and jealousy written all over his face.
"What the hell is going on here?" Oliver yelled, his fists clenched at his sides.
Natalie tried to play it cool, trying to act like nothing was out of the ordinary. "We were just talking, Oliver. The Baron and I are old friends."
But Oliver wasn't buying it. He knew that something was going on between Natalie and the Baron, and he wasn't about to let it continue.
"You can forget about this, Natalie. You're coming home with me right now," Oliver said, his voice laced with anger.
But Natalie wasn't ready to give up on the Baron just yet. She stood up from his lap and faced Oliver, her eyes blazing with determination.
"I'm not going anywhere with you, Oliver. I'm in love with the Baron," she said, her voice firm.
Oliver's eyes widened in shock. He had never imagined that his wife would choose a shark over him. But as he looked into Natalie's eyes, he knew that he had lost her.
"Fine, go be with your fishy lover. But just know that you're dead to me," Oliver spat out before storming out of the room.
Natalie watched him go, feeling a mix of sadness and relief. She knew that her life with the Baron wouldn't be easy, but she was ready to take on the challenges that lay ahead. She turned back to the Baron, who was watching her with a mixture of amusement and admiration, and gave him a sly smile.
"Well, that went well," she said with a laugh.
The Baron chuckled, his fins shaking with laughter. "I told you that your husband would never understand our love," he said, pulling Natalie back into his embrace.
As they sat together, surrounded by the midgets on their unicycles and the hobgoblins playing their instruments, Natalie knew that she had made the right choice. She was in love with the Baron, and nothing would ever tear them apart."
age 18-21 = Carter 45%. Reagan 44%
age 22-29 = Carter 44%, Reagan 44%
age 30-44 = Carter 38%, Reagan 55%
age 45-59 = Carter 39, Reagan 55%
age 60plus = Carter 41, Reagan 55%
So Reagan did win by landslide with voters 30 and older, but NOT with "all" ages voting in 1980.
https://ropercenter.cornell.edu/how-groups-voted-1980
However, Reagan DID win a landslide with all age groups in 1984.
https://ropercenter.cornell.edu/how-groups-voted-1984
This will simply accelerate the process.
However our leaders need to ensure we have enough electricity generation in place to plug the gap. It’s not something that seems to be a priority.
And now ChatGPT introduces a libidinous, dominant, margarine-smuggling great white shark called... The Baron
*stunned face emoji*
— On the first question: There’s a consensus among many House Republicans, one that few would dare utter publicly, that if McCarthy starts the 118th Congress as speaker, he’s not likely to end it that way.
If he’s able to lock down the 218 votes he needs to be speaker, the thinking goes, he likely will have given away the store to conservatives — including the “Never Kevin” crowd’s demand to make it easier to call a vote to oust the speaker. Many Republicans are already predicting the Freedom Caucus will use that tool, known as the “motion to vacate,” against McCarthy as soon as he strays from a conservative hard line. . .
. . . McCarthy allies tell us the real looming threat is the debt ceiling deadline. At some point next year — likely in the third quarter,according to the Bipartisan Policy Center — McCarthy will have to figure out a way to raise the federal debt limit or risk a national default, a move that will tear the GOP conference in two. Democrats are already signaling they won’t negotiate at all, let alone offer the draconian spending concessions the right flank is demanding. . . .
One possibility that some are eyeing as McCarthy’s get-out-of-jail-free card: a discharge petition signed by all House Democrats and a handful of moderate Republicans to raise the debt ceiling. Such a move would theoretically allow rank-and-file members to bring a debt increase to the floor without McCarthy’s blessing, allowing the GOP leader to rail against any deal while centrists take the heat.
— On the second question: House Republicans are also gossiping about what McCarthy would do if he fails a second time to win election as speaker. . . .
. . . . Many Republicans doubt he’d be interested in smoothing the way for another speaker.
As we’ve reported, No. 2 leader STEVE SCALISE is the most likely next in line. But McCarthy and Scalise have long had a tense relationship, dating back to when the Louisianan voiced his desire to lead the conference someday — comments McCarthy’s team took as a slight. (Scalise says he supports McCarthy for speaker and will not challenge him.)
Already some McCarthy allies are quietly grumbling that Scalise could do more to help settle the speaker race, and one senior GOP aide expressed a worry to us recently that if McCarthy falls short again, he’ll try to “sabotage” whomever emerges: “Kevin’s going to have to decide how to leave,” the person said. “Will he burn down the institution?” . . .
https://www.politico.com/newsletters/playbook/2022/12/29/two-burning-questions-about-kevin-mccarthy-00075787
As for Rishi, he is still polling better than his party- even if that is a late honeymoon artefact. Either he drags the party up, in which case well done him. Or more likely, the party drags him down, in which case he might as well carry the can for the impending defeat.
Charging infrastructure is a significant barrier to greater EV takeup: second only to the high cost of EVs.
As an aside, allegedly the Tesla Semi can take a 1Mwh charger. That'll cause problems for unprepared local grids.
No.
They were also early in the using batteries to back their chargers, to even out grid loading.
endex