For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Yes, it could be because one is not giving the right prompts, but why would one work as a writer to find the right prompts to get a fairly OK piece of text, not just use the prompts? The prompts would be far more succinct and useful to read than silly spiels of cod writing.
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts (it had a head wound full of maggots).
I think we are a little away from AI doing any of that.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
Google has a brilliant AI podcast tool, I'm afraid. We are starting to use it at one of the places where I do some work. When I first heard it I could not believe how good it was. And it is only going to get better, of course.
This is Notebook LM?
I think one big question is how Copyright will apply, and where the limitations will be drawn. For the UK, where does consumption and summarising of others' material fall under Fair Dealing, given that quotes and citations are limited to what is necessary.
We've been through this before. I'd draw a parallel with our press having routinely copied others' content to within an inch of what they think they can get away with along the line of the edge of enforcible law.
I'd suggest that the relationship between the actor and their AI bit will reflect that between Principal and Agent.
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts.
I think we are a little away from AI doing any of that.
‘We must shoot all deers to prevent similar incidents happening’
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
If I was your age in your job I would fiercely believe what you believe, because accepting what I’m saying means your career is going to end within 5-7 years or less. But it is nonetheless the case
I have experimented with ai a little in copywriting. Apart from writing in an incredibly repetitive style which it can't change, it has (as one would expect) a complete inability to add value to the information you give it. Tell it a car is red and ask for half a page, and it will produce half a page of annoyingly repetitive filler text but it will only impart that the car is red. One may as well just put the unvarnished prompt information out there in public - it will be a lot quicker and less annoying to read.
By the same token, is ai in totality giving us any more than we're putting in? We put in vast computing power and it's producing answers of varying quality - wrapped in the flash and bang of a message-like interface and the ability to fill a screen with words very quickly. We keep being told it will reach sentience any minute now, but we were being told that in the early days of ChatGPT.
I am not sure how AI gets and writes news stories.
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
How is that an issue though ? Compute power will only grown over time.
We're already at the stage where a directed AI can perform specific cognitive tasks at a level well beyond most people. In that sense doing what the machine age did for manual labour.
We've not really worked out what that equivalent means for cognitive work - and we're only a few years into the process. And compute advances far more quickly than did the development of mechanical systems.
Even assuming no conceptual breakthroughs (which is daft, as they will happen), it will remake economies and leave many of us stranded.
And it potentially concentrates power in the hands of a few, orders of magnitude beyond what machines did.
On a more cheerful note, I am halfway to getting a visa for Myanmar. Instead of being rejected outright, they are asking for absurd documents in PDF. I have never got this far before
*airpunch*
Aren't you off to Cornwall ?
And watch out for a revolution in Myanmar. It's not exactly stable at the moment.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Ukrainians update BTW... The older one has got a job as a performer on a cruise ship starting next year. I told her to lie about being able to speak excellent German to get the job so I've got about 5 months to teach myself then her German to at least B1 level. The younger one wants to be a primary school teacher and is pursuing that. Return to Kharkov is definitively not on the agenda.
Wouldn't spending six months a year on a cruise ship mess up their future immigration status? Can't do the five year route like that.
On a more cheerful note, I am halfway to getting a visa for Myanmar. Instead of being rejected outright, they are asking for absurd documents in PDF. I have never got this far before
*airpunch*
Aren't you off to Cornwall ?
And watch out for a revolution in Myanmar. It's not exactly stable at the moment.
I’ve had to delay my Cornish departure. A last bit of work. Eeek
Still hoping to get to Myanmar this winter. I don’t mind revolutions. I’ve seen a couple
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
If I was your age in your job I would fiercely believe what you believe, because accepting what I’m saying means your career is going to end within 5-7 years or less. But it is nonetheless the case
I have experimented with ai a little in copywriting. Apart from writing in an incredibly repetitive style which it can't change, it has (as one would expect) a complete inability to add value to the information you give it. Tell it a car is red and ask for half a page, and it will produce half a page of annoyingly repetitive filler text but it will only impart that the car is red. One may as well just put the unvarnished prompt information out there in public - it will be a lot quicker and less annoying to read.
By the same token, is ai in totality giving us any more than we're putting in? We put in vast computing power and it's producing answers of varying quality - wrapped in the flash and bang of a message-like interface and the ability to fill a screen with words very quickly. We keep being told it will reach sentience any minute now, but we were being told that in the early days of ChatGPT.
I am not sure how AI gets and writes news stories.
WRT writing, 99% of people who write don’t make a living from it. They do it because they like it. You have to be incredibly lucky to make a living out of it. I don’t see AI changing that.
Something like my Dissertation could not be done by AI as the amount of original research I did would be beyond its capacity. AI could certainly summarise secondary sources far more quickly than I ever could, but the work requires a lot of study of unpublished archival material.
The revised GDP figures for April to June down to 0.4% proved what many informed sources believed which was the "gangbusters" increase in Q1 was a statistical freak and persuaded Sunak to cut and run for early July.
The July to September figure PRE budget in October cannot be attributed as Mel Stride is trying to do to anyone other than the fallout from the disasterous Tory Government and one othe factor very conveniently ignored "the Farage Riots" when large swathes of major Cities and Towns were "no go" areas for 2-3 key weeks in the height of summer and the School holidays.
Labour have made mistakes in over egging the gloom and Reeves first budget will only be seen in a true light I would suggest in 2 or 3 years time as being the Foundation setter she suggests or not!
The first positive outcomes of the Labour October Budget will start to filter through from April.
The biggest fear is the Gilts market, almost wholly being impacted by turmoil in USA and Germany which will impact on 5-10 year UK yields at a time when the UK market had actually read, considered and reacted to the October UK Budget and largely discounted any negative reaction back to pre-October levels before Trumps reelection and now fears over that and Elections in Germany and turmoil and France.
Hopefully the investment that Labour obtained of over 60bn, the much anticiapted trip to Beijing, which will hopefully generate billions more and new more open relationship with China will offset the shit show likely to hit Washington from
Shenzhen88.
Tomorrow belongs to Xi
Heres a bet
Who gets a full State Visit first?
Trump-Musk
Xi
I hope it's Xi - he has far more to offer the UK and we have far less to fear from him and far more to gain!
The revised GDP figures for April to June down to 0.4% proved what many informed sources believed which was the "gangbusters" increase in Q1 was a statistical freak and persuaded Sunak to cut and run for early July.
The July to September figure PRE budget in October cannot be attributed as Mel Stride is trying to do to anyone other than the fallout from the disasterous Tory Government and one othe factor very conveniently ignored "the Farage Riots" when large swathes of major Cities and Towns were "no go" areas for 2-3 key weeks in the height of summer and the School holidays.
Labour have made mistakes in over egging the gloom and Reeves first budget will only be seen in a true light I would suggest in 2 or 3 years time as being the Foundation setter she suggests or not!
The first positive outcomes of the Labour October Budget will start to filter through from April.
The biggest fear is the Gilts market, almost wholly being impacted by turmoil in USA and Germany which will impact on 5-10 year UK yields at a time when the UK market had actually read, considered and reacted to the October UK Budget and largely discounted any negative reaction back to pre-October levels before Trumps reelection and now fears over that and Elections in Germany and turmoil and France.
Hopefully the investment that Labour obtained of over 60bn, the much anticiapted trip to Beijing, which will hopefully generate billions more and new more open relationship with China will offset the shit show likely to hit Washington from mid January
I would hope for Growth of 1.5% in 2025, rising to 2.5% in 2026 and some genuine hopes of a sustained and positive period afrer that.
As for the "R" word.....quite simply they can and will cause mischief, they can and will orchestrate civil disobedience and I have no doubt that by the summer of 2028 both America, Europe and the UK will look upon Farage / Trump and Musk and their ilk with the complete and utter disdain and horror that they deserve. Either Reform or the Tories are going to implode and the Tories only hope I would suggest is to ditch tgeir right wing and useless Leader and to form a new Tory Party in the mould of Major / Clarke / Heath....outward looking and eminently sensible conservatives with a small c. At that point they do become a genuine threat to Labour.
Starmer must hold his nerve, and I believe he will.
Shenzhen88.
Tomorrow belongs to Xi
Heres a bet
Who gets a full State Visit first?
Trump-Musk
Xi
I hope it's Xi - he has far more to offer the UK and we have far less to fear from him and far more to gain!
It will have to be Trump if SKS can get him to come.
Outside chance of Macron as the baleful swansong of radical centrist Europe.
No chance of Xi, and it would cause domestic ructions for Modi who has occasion to drape himself in the banner of anti-imperialism from time to time.
Fash Karen from Italy would tick a lot of boxes. European but a right wing shitbag. Close defense partner but slightly Ukro-skeptical.
Lab right up Trump’s booty, right down to their foot soldiers. How could Trump resist a ticker tape parade down Cromwell St.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
My 22 year old granddaughter has written several stories, is an avid reader of books, and describes herself as a 'writer'
I will not repeat your comments to her, no matter they could have some truth, as I would not want to upset her dreams and ambitions
There will absolutely always be opportunities. She should not give up.
She is a very determined young lady with 2025 being the year she gains her BA, having studied Italian language and culture
She could become a translator, not least a she speaks English, Welsh, Italian, French and is passable in Japanese, but she seems to want to transition to law
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
Google has a brilliant AI podcast tool, I'm afraid. We are starting to use it at one of the places where I do some work. When I first heard it I could not believe how good it was. And it is only going to get better, of course.
Well, I still hope you listen to mine. It's made by a human.
I suspect TTRPGs might be more resilient than most areas to AI as it's about interaction with other people. I'm hoping the desire for authenticity in podcasts might help them hold up too.
Well let's hope not but it has to be a material possibility, not just for us but for many of our peers too. I think the "recession or not" question is a bit overdone. A sluggish economy is still a sluggish economy even if it avoids two negative quarters in a row. The difference between -0.1% and +0.1% is not actually that meaningful. But let's hope not anyway.
Politicians really should switch their focus onto health and happiness rather than the economy.
Firstly as they are both more important. And secondly they are far easier for the government to make shift changes on.
And which as it happens would probably boost the economy. But I fear there's a touch of 'me and you vs the world' on this one.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Merry Christmas!
Our AI-overlords will never take Xmas from us!!!
The jobs which AI will find hardest to do are the stereotypically masculine ones (like bricklaying) and the stereitypically feminine ones (like nursing). Perhaps we are heading back to 1950s style relationships between the sexes.
https://x.com/yejinjgim/status/1871120204077715542 While former Defense Intelligence Commander Roh Sang-won is suspected of being involved in planning the martial law emergency on December 3, a memo suggesting "inducing North Korean attacks at the Northern Limit Line (NLL)" was found in his notebook...
..It was also discovered that the expression "shoot to kill" was written in former Commander Roh's notebook.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Merry Christmas!
Our AI-overlords will never take Xmas from us!!!
The jobs which AI will find hardest to do are the stereotypically masculine ones (like bricklaying) and the stereitypically feminine ones (like nursing). Perhaps we are heading back to 1950s style relationships between the sexes.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
My granddaughter will find her way and just enjoys writing and not for commercial reward
May I take this opportunity of wishing you and yours a very Happy Christmas
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
I’ve promised not to discuss AI any more this chrimbo. Someone else will have to answer!
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
If I was your age in your job I would fiercely believe what you believe, because accepting what I’m saying means your career is going to end within 5-7 years or less. But it is nonetheless the case
I have experimented with ai a little in copywriting. Apart from writing in an incredibly repetitive style which it can't change, it has (as one would expect) a complete inability to add value to the information you give it. Tell it a car is red and ask for half a page, and it will produce half a page of annoyingly repetitive filler text but it will only impart that the car is red. One may as well just put the unvarnished prompt information out there in public - it will be a lot quicker and less annoying to read.
By the same token, is ai in totality giving us any more than we're putting in? We put in vast computing power and it's producing answers of varying quality - wrapped in the flash and bang of a message-like interface and the ability to fill a screen with words very quickly. We keep being told it will reach sentience any minute now, but we were being told that in the early days of ChatGPT.
I am not sure how AI gets and writes news stories. So breaking news reporting and investigative journalism will probably survive - if people still want it. I doubt it will pay very well, though.
It seems pretty clear to me that copywriting is just about done for in the age of AI. Maybe 10% of copywriters will survive by turning AI-generated copy into something better. And that is the opportunity that AI does present - humans will always be able to polish it and there will be others who will find lucrative ways to interrogate and manipulate it. Generating prompts is going to become a significant skill.
As someone in a related industry, I think you're completely wrong about writing copy, based on an understandable misconception of what AI is and what it can do. There's a lot of product research that goes into it, which you then need to marshall into coherent prompts. Ai can then take this information and inflate it like expanding insulating foam into coherent (though dreadfully trite and samey) filler text. A very limited skillset and not one that is fundamentally important in explaining to someone why they should buy something.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Merry Christmas!
Our AI-overlords will never take Xmas from us!!!
The jobs which AI will find hardest to do are the stereotypically masculine ones (like bricklaying) and the stereitypically feminine ones (like nursing). Perhaps we are heading back to 1950s style relationships between the sexes.
No optimus robots could do bricklaying. The brickies will have to take drugs and play videogames all day.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I know I am testing the mods’ patience and I know that PB doesn’t like hearing this stuff (few do) so I’ll make this my last AI comment before Christmas
The “best” AIs at the moment are Large Language Models. So it is unsurprising that the first jobs they’re coming for are those most engaged with words and language - writers and translators
But all kinds of AI are coming - agentic etc - and they are brilliant in all cognitive domains and will only get better so ALL cognitive jobs are threatened, and most are surely doomed
Finally, if you’re using the best AIs but still getting shit output I suggest that’s because you’re shit at prompting
On a more cheerful note AI might solve climate change, save the planet, and cure ageing, so it’s not all bad
And on that festive happiness, Merry Christmas to all! xxx 🕺🥂🎅🍾
Merry Christmas!
Our AI-overlords will never take Xmas from us!!!
The jobs which AI will find hardest to do are the stereotypically masculine ones (like bricklaying) and the stereitypically feminine ones (like nursing). Perhaps we are heading back to 1950s style relationships between the sexes.
Mrs Flatlander is the bricklayer in this household, as I'm too fussy and take too long.
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
If I was your age in your job I would fiercely believe what you believe, because accepting what I’m saying means your career is going to end within 5-7 years or less. But it is nonetheless the case
If what you say is true most careers will be gone with a decade, certainly permanent jobs.
Making a UBI funded by a robot tax inevitable if AI leads to mass unemployment
Maybe they will just let us starve to death. Make it like the hunger games. We already know our establishment doesnt give a t.ss about us.
Prediction-wise there will be a technological development in 2025, (if it has not already happened in 2024 (or even 2023)) which will be so epochal we will look back on it and think “that’s when human life changed forever, that’s when our universe was over-turned”
However we might not realise until 2026. Or 2027, at a stretch
Unless you can pinpoint what you think it might be, that's just a silly prediction.
"Something massive will happen, but we won't know for years!!!!"
What was the last technological development that was world-changing, that occurred in just a year?
(2025 is the 200th anniversary of the Stockton and Darlington Railway; arguably (and they will argue...) the first passenger railway in the world. I'd argue that was the start of an epochal change to the world. Perhaps the biggest, as it changed the entire way society formed itself.)
Ok. I hope the mods will forgive me as it’s Christmas
I have to leave the site anyway for a few days to drive to Cornwall and see fam and friends for chrimble - so they don’t have to ban me
Here goes. I believe we have reached Artificial General Intelligence. A few days ago OpenAI launched ChatGPTo3. It performs better than 98% of Phd candidates in MULTIPLE domains - ie it is smarter than 99.996% of humanity. It is “the 175th best coder in the world”. It passed the Arc AGI test - a test so hard it was thought it would take until 2030 for any AI to have a chance, a test that was contrived - in part - as a skeptical challenge to show AI could never equal or surpass the best humans
Perhaps more importantly, in my own secondary field of writing I have seen - in the last year - AI going from “pretty good” at editing and writing to “OMFG that’s professionally good” - this is in one year
We are now on the exponential bit of the AI curve. GPTo3 came 3 months after GPTo1. The speed of progress is actually accelerating. Human inertia being what it is, this may take some time to impact, but if you have a job which involves a screen and cognition, you should expect your job to disappear within 5 years, or less
Here’s a graph to represent what has just happened
(Snip)
Yes, and I believe many such claims are b/s.
It literally passed the Arc AGI test
Also, three weeks ago I lost one of my main flint knapping jobs - to AI. It is now personally impacting
Look away, if you prefer
Yes, and I believe many such claims are b/s.
The current ML/AI tech is good. Good enough to fool many. But is it actually intelligence, let alone an AGI?
I'd argue the many stupid mistakes they make prove otherwise.
Had some good news in the last week of work: we’ve won a £1M grant for a project on medical AI. There are many great and wonderful things we can do with AI in healthcare. But, no, we are not near to AGI.
In the coding world - the fiddling of the benchmarks is mildly amusing.
“AI” is used as a code completion tool. The whole “95% of code written by” thing is also funny. Nearly that high a percentage of code is written by the autocomplete tools already.
Consider a simple bean class in Java - you write the names of a dozen parameters (say). The types are autocompleted from a letter or two, with a keyboard shortcut. Then the getters, setters, constructors, equals, hashCode, toString and Builder are auto-generated with a click. And formatted to look nice.
Class is probably 3 figures of lines. Human partially wrote a dozen. No “AI” involved.
“AI” adds a layer on that, but that simply pushes the up the level of “boiler plate” coding.
In summary:
I don't know enough about Java, but in translation the AI will get 90% right and will get 10% wrong in a mostly predictable way - typically confusing positives to negatives and vice versa due to word order or complex sentences. However, there are still about 3-5% errors that are unpredictable - e.g. a whole clause simply ignored. That has made translation for the EU into a win-win for both EU and translators - the pay is down by 25%, productivity is up by perhaps 60%, and the prospect of eliminiating the human element entirely is very low, because the consequences of overlooking a mistake are potentially huge.
For casual translation, though - getting the sense of a piece with 90% accuracy - the need for human intervention has more or less disappeared. Curiously, the number of people doing serious translation where a mistake is a Big Deal seems to have dropped - I'm given first refusal on Danish to English translation by several agencies, and even German to English seems to have demand exceeding supply.
Useful Anecdote
Useful Data:
“AI threatens to decimate the translation profession, according to a new survey by a British union.
Almost four in ten translators (36%) said they’ve already lost work due to generative AI. Nearly half of them (43%) said the tech has decreased their income.
They fear far worse is to come. Over three-quarters of translators (77%) believe GenAI will negatively impact future income from their creative work.”
I don't disagree. The reason why EU human translation based on AI first drafts remains robust is that the jump to relying totally on AI for LEGISLATION is scary. For a guide to good places to visit, or anything like that, AI now rules. But if you can be sued for £X million, or simply cause £X million in lossses, if you get legislation wrong, it doesn't seem safe to cut out the human check. Obviously the human isn't perfect either, but human error in translation tends to be different - typically leaving something out, rather than getting it flatly wrong. Having a human step seems a sensible precaution for the forseeable future.
An interesting question is whether AI translation of fiction is likely to be better or worse. I'd think that it will continue to improve and become clearly better than human translation.
Literary translation is incredibly subjective, isn't it? You can see AI providing the first draft and then a human doing the final one, taking it to places that individual alone was capable of imagining. Umberto Eco said the English translation of The Name of the Rose was better than the original in Italian. A human did that.
It was very hard to believe that the Name of the Rose was not written in English. In my experience, though, that was a rare achievement.
AI will never match the human capacity to devise pointless regulations, ever more complicated types of compliance, and more complex procedures, in order to keep a whole class of people in employment.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
I’ve promised not to discuss AI any more this chrimbo. Someone else will have to answer!
It doesn't surprise me that it was an American magazine - the ai 'style' it can't shed is developed along the lines of a middle of the road American blog. If one intended to replace a full-time journalist with ai, even if they were writing crappy advertorial or Top 10 lists, one would at least need an additional day of editing support to check its work and amend some of the more egregious ai-isms, and then someone to prompt it, replacing a full time worker with a part time one - unfortunate for the worker, but not a huge drama, and one in my view that could be liable to eventual reversal.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
I’ve promised not to discuss AI any more this chrimbo. Someone else will have to answer!
It doesn't surprise me that it was an American magazine - the ai 'style' it can't shed is developed along the lines of a middle of the road American blog. If one intended to replace a full-time journalist with ai, even if they were writing crappy advertorial or Top 10 lists, one would at least need an additional day of editing support to check its work and amend some of the more egregious ai-isms, and then someone to prompt it, replacing a full time worker with a part time one - unfortunate for the worker, but not a huge drama, and one in my view that could be liable to eventual reversal.
I lied about it being American so as to disguise its identity
AI will never match the human capacity to devise pointless regulations, ever more complicated types of compliance, and more complex procedures, in order to keep a whole class of people in employment.
If that were true we should be handing over regulation to them immediately.
A building story in our area is the number of care homes that are shutting, especially the smaller ones. This is mainly being driven by the inability to provide anything like the level of staffing, training, monitoring and form filling needed for the sort of money that local authorities can afford to pay. The care homes also have to fund the regulators that are putting them out of business.
One home that we approached about my Mother in law would only accept new residents who could prove they had 2 years worth of fees in the bank. They were not willing to accept the risk of a new resident who would only be producing the local authority rate after a relatively short period. Any home with a majority of authority paid residents is facing closure or prosecution. It is not a great choice.
It is also why we have the most expensive child care in the world making it very hard for many parents to find work that covers the cost.
I suppose the question is how you define a 'bad candidate'. Kerry for what it's worth wasn't a poor candidate, he came within one state of beating an incumbent president who still retained significant sympathy amid the War On Terror. Won the debates - and may have actually been truly made toast by the Bin Laden tape - something he had no control over. He was almost certainly the best pick available to the Democrats that year.
What Kerry wasn't of course was an Obama or Reagan - a candidate with the political skills to bend a race their way by force of personality. But those are rare, generational figures. Of conventional politicians you could add Clinton (Trump rather sits in his own category) - but he did get lucky with third party candidates and his opponents. Otherwise it goes along with the fundamentals and we tend to malign candidates who lost as "bad" when they may have actually been middling or in some cases decent, but defeated by the headwinds.
Harris I think sits in that middling area - she didn't have the skills to dig her party out of the hole they'd dug by hanging on to Biden, but not a bad one who made obvious screw ups that affected the race. Plus one with a profile that was unhelpful in terms of being a Trump-killer. Basically, as with Biden in 2020, you needed to select the candidate with the most appeal to the Mid-west, as that's the entire ball game - and could've saved the Democrats even as the national picture shifted Trump's way.
But to say was a "bad candidate" probably lets the Democrats off the hook when it wasn't particularly anything she did that lost them the election - in fact she improved their position, just not by enough. When they should really be going back over the past 4 years and examining the decisions that put them in such a difficult place that Harris wasn't able to magic away with force of personality.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
I’ve promised not to discuss AI any more this chrimbo. Someone else will have to answer!
It doesn't surprise me that it was an American magazine - the ai 'style' it can't shed is developed along the lines of a middle of the road American blog. If one intended to replace a full-time journalist with ai, even if they were writing crappy advertorial or Top 10 lists, one would at least need an additional day of editing support to check its work and amend some of the more egregious ai-isms, and then someone to prompt it, replacing a full time worker with a part time one - unfortunate for the worker, but not a huge drama, and one in my view that could be liable to eventual reversal.
I lied about it being American so as to disguise its identity
Ah. So the Speccie is now using AI under Gove. What's the column? Come on. You can tell us. You're amongst friends. We'll be discreet.
AI will never match the human capacity to devise pointless regulations, ever more complicated types of compliance, and more complex procedures, in order to keep a whole class of people in employment.
I'm not so sure. It's science fiction, but if you go back to RUR, the 1920 play by Karol Capek that coined the term robot, the robots, having slaughtered the human race, decide keep themselves busy by building pointless houses.
If we outsourced regulation to AI, I can see the AI deciding to exterminate the human race through rules banning procreation and then continuing to even more pointless regulations just to keep itself busy.
Maybe I'll write a play about it?
As a homage, I think I'd call the main AI character Sir Humphrey ...
I suppose the question is how you define a 'bad candidate'. Kerry for what it's worth wasn't a poor candidate, he came within one state of beating an incumbent president who still retained significant sympathy amid the War On Terror. Won the debates - and may have actually been truly made toast by the Bin Laden tape - something he had no control over. He was almost certainly the best pick available to the Democrats that year.
What Kerry wasn't of course was an Obama or Reagan - a candidate with the political skills to bend a race their way by force of personality. But those are rare, generational figures. Of conventional politicians you could add Clinton (Trump rather sits in his own category) - but he did get lucky with third party candidates and his opponents. Otherwise it goes along with the fundamentals and we tend to malign candidates who lost as "bad" when they may have actually been middling or in some cases decent, but defeated by the headwinds.
Harris I think sits in that middling area - she didn't have the skills to dig her party out of the hole they'd dug by hanging on to Biden, but not a bad one who made obvious screw ups that affected the race. Plus one with a profile that was unhelpful in terms of being a Trump-killer. Basically, as with Biden in 2020, you needed to select the candidate with the most appeal to the Mid-west, as that's the entire ball game - and could've saved the Democrats even as the national picture shifted Trump's way.
But to say was a "bad candidate" probably lets the Democrats off the hook when it wasn't particularly anything she did that lost them the election - in fact she improved their position, just not by enough. When they should really be going back over the past 4 years and examining the decisions that put them in such a difficult place that Harris wasn't able to magic away with force of personality.
Maybe if Harris had had a longer run at it…….. Did Biden hang too long to give her a reasonable shot?
I suppose the question is how you define a 'bad candidate'. Kerry for what it's worth wasn't a poor candidate, he came within one state of beating an incumbent president who still retained significant sympathy amid the War On Terror. Won the debates - and may have actually been truly made toast by the Bin Laden tape - something he had no control over. He was almost certainly the best pick available to the Democrats that year.
What Kerry wasn't of course was an Obama or Reagan - a candidate with the political skills to bend a race their way by force of personality. But those are rare, generational figures. Of conventional politicians you could add Clinton (Trump rather sits in his own category) - but he did get lucky with third party candidates and his opponents. Otherwise it goes along with the fundamentals and we tend to malign candidates who lost as "bad" when they may have actually been middling or in some cases decent, but defeated by the headwinds.
Harris I think sits in that middling area - she didn't have the skills to dig her party out of the hole they'd dug by hanging on to Biden, but not a bad one who made obvious screw ups that affected the race. Plus one with a profile that was unhelpful in terms of being a Trump-killer. Basically, as with Biden in 2020, you needed to select the candidate with the most appeal to the Mid-west, as that's the entire ball game - and could've saved the Democrats even as the national picture shifted Trump's way.
But to say was a "bad candidate" probably lets the Democrats off the hook when it wasn't particularly anything she did that lost them the election - in fact she improved their position, just not by enough. When they should really be going back over the past 4 years and examining the decisions that put them in such a difficult place that Harris wasn't able to magic away with force of personality.
Maybe if Harris had had a longer run at it…….. Did Biden hang too long to give her a reasonable shot?
Dunno. It seems the more people saw of her the more they disliked her?
A longer run-in might have seen Trump win by an even bigger margin?
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
Look at the graph I posted. No one believes what you’re saying
The issue is compute power, Blackwell doesn't look like a big leap even with the giant chip and while there have been advancements it's nothing like going from the old Google computer vision to midjourney. Those big leaps just aren't available now because we've already achieved them.
If I was your age in your job I would fiercely believe what you believe, because accepting what I’m saying means your career is going to end within 5-7 years or less. But it is nonetheless the case
I have experimented with ai a little in copywriting. Apart from writing in an incredibly repetitive style which it can't change, it has (as one would expect) a complete inability to add value to the information you give it. Tell it a car is red and ask for half a page, and it will produce half a page of annoyingly repetitive filler text but it will only impart that the car is red. One may as well just put the unvarnished prompt information out there in public - it will be a lot quicker and less annoying to read.
By the same token, is ai in totality giving us any more than we're putting in? We put in vast computing power and it's producing answers of varying quality - wrapped in the flash and bang of a message-like interface and the ability to fill a screen with words very quickly. We keep being told it will reach sentience any minute now, but we were being told that in the early days of ChatGPT.
I am not sure how AI gets and writes news stories.
It reads twitter and writes a summary.
So the main news story of the day involves cats wearing Santa outfits.
I suppose the question is how you define a 'bad candidate'. Kerry for what it's worth wasn't a poor candidate, he came within one state of beating an incumbent president who still retained significant sympathy amid the War On Terror. Won the debates - and may have actually been truly made toast by the Bin Laden tape - something he had no control over. He was almost certainly the best pick available to the Democrats that year.
What Kerry wasn't of course was an Obama or Reagan - a candidate with the political skills to bend a race their way by force of personality. But those are rare, generational figures. Of conventional politicians you could add Clinton (Trump rather sits in his own category) - but he did get lucky with third party candidates and his opponents. Otherwise it goes along with the fundamentals and we tend to malign candidates who lost as "bad" when they may have actually been middling or in some cases decent, but defeated by the headwinds.
Harris I think sits in that middling area - she didn't have the skills to dig her party out of the hole they'd dug by hanging on to Biden, but not a bad one who made obvious screw ups that affected the race. Plus one with a profile that was unhelpful in terms of being a Trump-killer. Basically, as with Biden in 2020, you needed to select the candidate with the most appeal to the Mid-west, as that's the entire ball game - and could've saved the Democrats even as the national picture shifted Trump's way.
But to say was a "bad candidate" probably lets the Democrats off the hook when it wasn't particularly anything she did that lost them the election - in fact she improved their position, just not by enough. When they should really be going back over the past 4 years and examining the decisions that put them in such a difficult place that Harris wasn't able to magic away with force of personality.
Maybe if Harris had had a longer run at it…….. Did Biden hang too long to give her a reasonable shot?
Yes, way too long. Had he stepped back before the primaries, she'd either have been a far more credible candidate in the general, or someone diss would have won the nomination.
Seriously, if you’re young and fancy a career in writing, give up. Unless you focus on memoir or war reportage, something AI might never do, it’s done. It’s over. There won’t be a career
I am intrigued by this notion. I have access to the newest paid ChatGPT and it's utterly shite at writing. I suppose you must have access to far better models.
What’s the point in my lying about the death of my profession? Seriously? I like attention but I don’t like it that much
Imparting this news does not make me happy. Quite the fucking opposite. I hate it (even if I find the underlying tech exciting). But it is happening and it is futile if not dangerous to deny it
I don't think you're lying for a second, I think you're wrong, having seen something impressive and drawn erroneous and exaggerated conclusions.
Ok I have to reply to that
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
Hence my point. How will ai help ordinary people. Sure its great for the 1% class but everyone else?
I’ve promised not to discuss AI any more this chrimbo. Someone else will have to answer!
It doesn't surprise me that it was an American magazine - the ai 'style' it can't shed is developed along the lines of a middle of the road American blog. If one intended to replace a full-time journalist with ai, even if they were writing crappy advertorial or Top 10 lists, one would at least need an additional day of editing support to check its work and amend some of the more egregious ai-isms, and then someone to prompt it, replacing a full time worker with a part time one - unfortunate for the worker, but not a huge drama, and one in my view that could be liable to eventual reversal.
I lied about it being American so as to disguise its identity
Ah. So the Speccie is now using AI under Gove. What's the column? Come on. You can tell us. You're amongst friends. We'll be discreet.
Now now. No need for naughtiness
Moving away from AI itself to the economics of the future. You only have to crunch the maths to see what’s coming
Imagine a company making creative widgets. Let’s call them artburgers. It can employ a team of ten artburger designers each on £100 a day, each designing one art burger a day. Or it can get a droid to churn out those ten artburgers a days basically for free - and sack nine artburger designers and keep one designer to polish up the ones made by the droid
Unfortunately for Artburger Inc the new mechanised artburgers aren’t as good as the man made ones. But then Artburger Inc makes the happy discovery that the consumers don’t care about the difference, indeed - they don’t even notice it. Only the pro human designers can spot the difference (and in the end that difference will disappear)
So Artburger Inc can now save £900 a day, £234,000 a year by sacking 90% of its designers. It will absolutely do that, not least coz its competitors will do the same
I am interested in the question about translation of fiction by AI. As is well known, translation of fiction is not just word for word translation. A classic example might be the Iliad. I ave half a dozen or more different translations ranging from Pope in the 18th Century, to the sublime E.V. Rieu and on to more modern translations by Fagles or McCrorie. Each is unique and each brings an entirely different feel to the work.
Will different AI's have different feels'. Will the same AI have a different 'feel' every time it is used? Does the fact it is AI remove the human emotion and render it pointless?
How is that an issue though ? Compute power will only grown over time.
That's how it's always been up till now, but it's getting harder and harder to find more performance. Silicon based semiconductors have almost run their course, new process nodes are offering small (often single-digit percentage) increases in performance and reductions in power draw with the draw-back of steeply rising cost per wafer produced.
You can see this having an effect in almost every market. In the two year gap between their Ryzen 7000 and 9000 processors AMD managed to get an average performance uplift of just 5%. NVidia is resorting to software-based tricks like frame generation to boost performance of their GPUs at the cost of image quality. Their upcoming 5000-series graphics cards are rumoured to be using an 'AI' texture compression system, again trading image quality for performance.
At the moment LLM based AI simply uses too much hardware and consumes too much power, the solution to that is not likely to come from better hardware.
Bonkers, and there's also Canada as the 51st State, seizing the Panama Canal, and invading Mexico to fight the drug cartels. Trump is touting stuff every bit as crazy as Putin's imperial ambitions.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Your second item interests me. Does AI do requirements capture? In my experience that's always been a place where we humans fall down quite badly, which leads to systems being developed that are the developers' idea of what the customer was asking for but not what was actually asked for. I've also encountered quite a lot of humans who simply don't believe that the customer could possibly want what they asked for.
It's one thing helping the customer to understand what they want and how it can be achieved (or why it can't be achieved in the way they hoped), but when you get developers deciding that the customers don't know what they're talking about in their own business it's a big problem.
Chump's trying to behave like a late-Victorian, when the USA just started, or manufactured, a war (ish) if they wanted anything near the N/S American continents.. He was born in 1946, at Junior/Senior School (whatever the Yanks call it) in the 1950s when it was Communism and McCarthy.
He's imposing what he thinks he knows, rather than bothering to find out about the things he doesn't know.
It will hit a problem imo when he finds out that the USA is not as hegemonic as it arguably was just after WW2.
I am interested in the question about translation of fiction by AI. As is well known, translation of fiction is not just word for word translation. A classic example might be the Iliad. I ave half a dozen or more different translations ranging from Pope in the 18th Century, to the sublime E.V. Rieu and on to more modern translations by Fagles or McCrorie. Each is unique and each brings an entirely different feel to the work.
Will different AI's have different feels'. Will the same AI have a different 'feel' every time it is used? Does the fact it is AI remove the human emotion and render it pointless?
I have lots of questions but no answers.
There's an author on YouTube who goes by the name "The Nerdy Novelist". He does a lot of work with LLM's, comparing different models, techniques, what this one is good/bad at. I just watch it out of curiosity more than for the actual 'writing' part of things - but it's interesting to me to see how people who professionally care about words are trying out all these tools and techniques, seemingly without the 'mainstream' of AI/LLM enthusiasts noticing their findings. There's also the 'Future Fiction Academy' who even have their own LLM writing app https://futurefictionacademy.com/raptor-write/.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
I've tried Copilot about ten times for IT related matters. On two occasions it gave a useful response; the other eight responses were completely non-useful. Has anyone else achieved a better success ratio?
I use Copilot constantly (literally at least once a day) via Microsoft Edge. I look thru the sources it gives and check it. That's why I gave a revised reply to MattW regarding political party structure: it missed that the Conservative Party isn't an unincorporated association and I hadn't read Conservatives vs Burrell well enough as I had to leave the flat to avoid the carol singers. On my return I checked more thoroughly and found the error.
I treat its replies as hints rather than deterministic replies: most of them are "good enough", but many havce errors and some have fatal errors. I don't have a ratio for you I'm afraid.
But guided by my conscience and my experience as a public defender, chairman of the Senate Judiciary Committee, vice-president, and now President, I am more convinced than ever that we must stop the use of the death penalty at the federal level.
So why not commute all 40 federal death sentences? I think it's very odd to support the death penalty depending on motivation.
I think that's an interesting shift at the edge of USA attitudes, to more pro-life policies. It's quite the reverse ferret on Joe Biden's part, from his stance in 1994:
As a senator, Biden championed a 1994 crime bill that expanded the federal death penalty to cover 60 new offences. He boasted: “I am the guy who put these death penalties in this bill.” The legislation is now widely seen as having contributed to mass incarceration, particularly affecting Black men, and many of those currently on death row were sentenced under its provisions.
And it will limit one tiny aspect of Trump's possibility for gesture-politics:
Under Trump, more people incarcerated in the federal system were put to death than under the previous 10 presidents combined.
If a man cannot change his mind after 30 years, when can he?
But, now I think of it, the Labour base would clearly view Streeting as being on the Right, so I'd probably want 16/1 or 18/1
Something like that.
I don’t think Streeting will be the next leader, unless there is a coronation, which I doubt.
If Starmer fails, the party will look leftwards, similar to how the Tory Party looked to the candidate of the right when they held their contests.
Depends if Streeting polls as the only leader who might retain Labour's majority or not.
On the Tory Party of the last 5 leaders only 2 were the more rightwing of the final 2, Boris and Truss, while the other rightwingers in the last 2 ie Jenrick, Leadsom and Davis were defeated. Truss ended up being replaced by Sunak before she could even fight a general election.
Whereas of the last 3 Labour leaders 2 were the more leftwing of the last 2 ie Ed Miliband and Corbyn and Starmer the only moderate, though even he is more of a Brownite than Blairite whereas Streeting is more true New Labour
People forget Gordon Brown was New Labour. Your implication might be valid though, since Brown was on the right of the party whereas Blair seemed more loosely attached.
The system needs pulling down. EHCP's have been a failure. We need more inclusion and an end to the division of SEN/non-SEN. Increasingly just a form of exclusion and exclusivity.
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
I've tried Copilot about ten times for IT related matters. On two occasions it gave a useful response; the other eight responses were completely non-useful. Has anyone else achieved a better success ratio?
It's something of a learning curve. I'm finding it increasingly useful as I come to understand more the sort of areas in which it can be useful. It's great for boilerplate, though it does make annoyingly sneaky mistakes sometimes. It's not so good at explaining concepts, frequently directly contradicting itself from one response to the next.
Copilot is great for data science/analytics which sort of makes sense. I've found it useful when writing regression models and slightly more complicated ones too. The key is to have a very well structured source tables, without that I've found copilot to be more miss than hit for my use cases.
It’s been a complete miss in my use cases - sadly they are the ones attached to other MS products that they want to hard sell copilot with
Give Cursor a go, I've heard it's well ahead of copilot.
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
I've tried Copilot about ten times for IT related matters. On two occasions it gave a useful response; the other eight responses were completely non-useful. Has anyone else achieved a better success ratio?
It's something of a learning curve. I'm finding it increasingly useful as I come to understand more the sort of areas in which it can be useful. It's great for boilerplate, though it does make annoyingly sneaky mistakes sometimes. It's not so good at explaining concepts, frequently directly contradicting itself from one response to the next.
Copilot is great for data science/analytics which sort of makes sense. I've found it useful when writing regression models and slightly more complicated ones too. The key is to have a very well structured source tables, without that I've found copilot to be more miss than hit for my use cases.
It’s been a complete miss in my use cases - sadly they are the ones attached to other MS products that they want to hard sell copilot with
Give Cursor a go, I've heard it's well ahead of copilot.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Your second item interests me. Does AI do requirements capture? In my experience that's always been a place where we humans fall down quite badly, which leads to systems being developed that are the developers' idea of what the customer was asking for but not what was actually asked for. I've also encountered quite a lot of humans who simply don't believe that the customer could possibly want what they asked for.
It's one thing helping the customer to understand what they want and how it can be achieved (or why it can't be achieved in the way they hoped), but when you get developers deciding that the customers don't know what they're talking about in their own business it's a big problem.
Good afternoon, everybody.
My experience on requirements is almost the opposite of yours - the problems in many of my projects have been where the customer got exactly what they asked for, which turned out not to be what they wanted. I wonder if AI can also make suggestions where things might not work as envisaged?
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
I've tried Copilot about ten times for IT related matters. On two occasions it gave a useful response; the other eight responses were completely non-useful. Has anyone else achieved a better success ratio?
It's something of a learning curve. I'm finding it increasingly useful as I come to understand more the sort of areas in which it can be useful. It's great for boilerplate, though it does make annoyingly sneaky mistakes sometimes. It's not so good at explaining concepts, frequently directly contradicting itself from one response to the next.
Copilot is great for data science/analytics which sort of makes sense. I've found it useful when writing regression models and slightly more complicated ones too. The key is to have a very well structured source tables, without that I've found copilot to be more miss than hit for my use cases.
It’s been a complete miss in my use cases - sadly they are the ones attached to other MS products that they want to hard sell copilot with
Give Cursor a go, I've heard it's well ahead of copilot.
How is that an issue though ? Compute power will only grown over time.
That's how it's always been up till now, but it's getting harder and harder to find more performance. Silicon based semiconductors have almost run their course, new process nodes are offering small (often single-digit percentage) increases in performance and reductions in power draw with the draw-back of steeply rising cost per wafer produced.
You can see this having an effect in almost every market. In the two year gap between their Ryzen 7000 and 9000 processors AMD managed to get an average performance uplift of just 5%. NVidia is resorting to software-based tricks like frame generation to boost performance of their GPUs at the cost of image quality. Their upcoming 5000-series graphics cards are rumoured to be using an 'AI' texture compression system, again trading image quality for performance.
At the moment LLM based AI simply uses too much hardware and consumes too much power, the solution to that is not likely to come from better hardware.
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts.
I think we are a little away from AI doing any of that.
‘We must shoot all deers to prevent similar incidents happening’
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Your second item interests me. Does AI do requirements capture? In my experience that's always been a place where we humans fall down quite badly, which leads to systems being developed that are the developers' idea of what the customer was asking for but not what was actually asked for. I've also encountered quite a lot of humans who simply don't believe that the customer could possibly want what they asked for.
It's one thing helping the customer to understand what they want and how it can be achieved (or why it can't be achieved in the way they hoped), but when you get developers deciding that the customers don't know what they're talking about in their own business it's a big problem.
Good afternoon, everybody.
It only does the requirements in very broad brush strokes. The end-users of our apps are very much office admins - as in their world is Excel and they know which buttons to press in the in-house apps. But you couldn't ask them much in the way of detail without making them frustrated. So our case is just 'capture the rough gist, tidy it up, pass it to a human'.
I saw this appear just the other week https://gitingest.com/ and am considering using it to give the LLM even more context, so it can add much better breakdowns to the developers. Luckily our developers are pretty good at understanding user requests - and if there's a 'whuuu?' will go back to the user to clarify.
I don't see any particular road-blocks in making it more fleshed out in other directions though. It's effectively free at our scale. A whole to-and-fro with the user and LLM is a fraction of a fraction of a penny.
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts.
I think we are a little away from AI doing any of that.
‘We must shoot all deers to prevent similar incidents happening’
Deer not deers
There is a warning sign for cyclists in my local park asking them to slow down because of Joggers disabled people deer
The last one has been altered to read "old deers".
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
It’s not up to them. It’s up to people on random bulletin boards.
Hah
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
One last remark on AI and the OpenAI announcements
1. Remember that other companies - esp Google, and to a lesser extent X - are CATCHING UP
2. Remember that robotics is now joining the same exponential surge
3. Think about this: GPTo3 is the “175th best coder” in the world. By ELO ranking (yes, adapted from chess) there is one - just one - coder at OpenAI better than o3 itself. So why don’t they get a bunch of o3’s to work on o3 itself? That would be the world’s best coding team. o3 would rapidly improve, then even more rapidly improve, then explode
That would and will be the recursive self improvement path to Artificial SUPER Intelligence. As predicted by some sci-fi writers years ago
It is now coming true. I think that deserves a Festive
BRACE
Right, now coffee, Cornwall, Christmas
🥂🎅💥
Surely one thing is true, once AI achieves "general intelligence" it very very quickly becomes better than humans can possibly be ?
The top chess programs are ranked over 3,000 ELO; way above any human.
Yes. The analogy is pretty good. Using ELO is clever
In this last year I have watched AI go from writing “pretty good articles about flint knapping but lol you can tell it’s AI why does it keep saying “delve”’ to “*stunned silence* fuck, was that done by AI? That’s genuinely good”
That’s in one year. In one tiny niche. Writing about flint sex toys. Good enough for me to lose a consultancy job
And we are on an exponential curve. So in a way whether it’s AGI or not doesn’t matter. There will always be some twat saying “but look it can’t cook a Mexican breakfast while patting a Labrador so it’s not AGI”. The dim and the fearful will always move the goalposts
Fact is AI will surge past human capabilities and likely soar away in the coming months and years
I don't think we're on the exponential part of the curve any longer. We're very much in the last 20% of delivery which will take much longer to achieve than the previous 80% we've had in the last 3-4 years.
I've tried Copilot about ten times for IT related matters. On two occasions it gave a useful response; the other eight responses were completely non-useful. Has anyone else achieved a better success ratio?
It's something of a learning curve. I'm finding it increasingly useful as I come to understand more the sort of areas in which it can be useful. It's great for boilerplate, though it does make annoyingly sneaky mistakes sometimes. It's not so good at explaining concepts, frequently directly contradicting itself from one response to the next.
Copilot is great for data science/analytics which sort of makes sense. I've found it useful when writing regression models and slightly more complicated ones too. The key is to have a very well structured source tables, without that I've found copilot to be more miss than hit for my use cases.
It’s been a complete miss in my use cases - sadly they are the ones attached to other MS products that they want to hard sell copilot with
Give Cursor a go, I've heard it's well ahead of copilot.
You get two weeks for free - then you're just bumped down to slow-mode.
It's really quite impressive. I don't use most of it's features - but the raw editing experience is quite something. They've really put a lot of effort into making it understand what you're trying to do - rather than just guess at a likely autocomplete a-la copilot.
A genuinely interesting conversation about AI translation this morning.
A few weeks ago I was contemplating a header arguing why it will take 50-100 years for Syria to get back to anything other than an uneasy peace, and thinking about different communities / groups understanding each other, and how long that takes just as a starter in a peace process.
One analogy I was thinking about for understanding and misunderstanding was the use of "unicorn" in the King James Bible Isaiah 34:14, and how the translation differs according to the values of the translator, the target community / audience, and the type of translation intended (study, for non-native speakers, to read aloud, literal, aimed at concepts or underlying meaning not precise language etc.)
The nearest term is probably something around wild ox, but terms used in English translations include the strange "unicorn" (which then leaks from the KJV into for example The Book of Mormon, as the writer rather fetishises the King James Bible), "rhinoceros", "wild buffalo" (USA, I assume) and others. That is then overlaid with a process by some modern-speculators trying to read the word-choice as an indication of classical mythology as a source for the original documents to justify their New Age beliefs.
Greg Wallace got a look in, because the same KJV also includes "satyrs", half-clothed man-goats.
Understanding these complications is one challenge for AI. Quality checking it may be more challenging.
There is a lovely table, which I cannot find again, of the different words. My photo quota today is how one place illustrates "rhinoceros".
Re Cursor - All it is is a VS Code fork that is plugged into ChatGPT and Claude. They recently purchased SuperMaven which has SOTA auto-complete code performance, which I presume will give them some moat over the 27,000 other VS Code forks that are plugging into ChatGPT / Claude.
Codeium is another one that people recommend. I think you know the market might be getting to frothy when a VS Code fork + some filtering of existing LLM gives you so much f##k up start up money you can make the ad for your IDE as a spoof on Silicon Valley tv show and hire one of the actors from that show.
What are the chances of Labour going sub-20 in a poll next year?
Looking at the already grim economic news allied with their total inteptitude, their complete lack of ideas, the arrival of Trump, the slowdown in Europe, the horror that is Starmer and Reeves, and the rise of Farage stealing Labour voters, I reckon there's a 40-50% chance
And once you go that low it gets pretty hard to come back
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts.
I think we are a little away from AI doing any of that.
‘We must shoot all deers to prevent similar incidents happening’
Deer not deers
There is a warning sign for cyclists in my local park asking them to slow down because of Joggers disabled people deer
The last one has been altered to read "old deers".
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
It’s not up to them. It’s up to people on random bulletin boards.
Hah
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
Probably I'm alone in PB. on that
That would be like wanting to join Italy provided we could still have regulations applied properly regardless of who you happen to know, could eat pasta where the sauce has been added after cooking rather than during, could order a cappuccino after 11 am, didn’t have to dress up in our best clothes and wander about the town every evening, and didn’t have to listen to whatever our mother is telling us.
Yes, we could do that, but we wouldn’t be Italian.
It’s not up to them. It’s up to people on random bulletin boards.
Hah
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
Probably I'm alone in PB. on that
That would be like wanting to join Italy provided we could still have regulations applied properly regardless of who you happen to know, could eat pasta where the sauce has been added after cooking rather than during, could order a cappuccino after 11 am, didn’t have to dress up in our best clothes and wander about the town every evening, and didn’t have to listen to what our mother is telling us.
Yes, we could do that, but we wouldn’t be Italian.
Yeah, but we'd be allowed to live in Italy
I don't want to be American but I would love the freedom to travel and live anywhere in the USA, the American protection of Free Speech, and the dynamism of the American economy
And a nutter like Trump might just be willing to offer it for the glory of adding a large new state to the USA (or four states?), and 70m people, and a mighty world city, and great universities, and some stunning countryside, and the English Premier League. And Wimbledon. And scones jam and cream. Etc
That would instantly make Trump one of the great American presidents, and he's not known for disliking praise and fame
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
And your biggest weakness?
I sometimes become a little too intense in my constant striving for perfection.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
You could pick that up at the interview, surely? Tell us what a great team player you are by taking personal credit for everything your team accomplished in the last five years.
Done my good deed for the day and another first for me - helped rescue a sheep tangled in brambles. Only concerning thing is the last animal I helped rescue, a deer, was then shot by the deer rescue charity that turned up after my efforts.
I think we are a little away from AI doing any of that.
‘We must shoot all deers to prevent similar incidents happening’
Deer not deers
There is a warning sign for cyclists in my local park asking them to slow down because of Joggers disabled people deer
The last one has been altered to read "old deers".
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
And your biggest weakness?
I sometimes become a little too intense in my constant striving for perfection.
That is the "learning lessons" of job interview answers....
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
Pretty certain. There are a few keywords that are give-aways to the very low-effort people, and there is an 'over-flowery' style. If you were just reading one CV you might not spot it, but when you have 30 with exactly the same patterns in it you begin to spot them.
Also the bargain-basement 'jazz up my CV' tools seem to just *love* adding nonsensical metrics and KPI's. "I improved the API response by 15% leading to an increase in user satisfaction of 37%" without any context. Just 'the API', 'user satisfaction' can be any plausible sounding thing. Again, if you were just reading one CV it might just roll past your eyes, but CV after CV after CV with the same lines.
We take for granted that the holidays to the sun we take come with cheaper prices. Holidaying somewhere more expensive than your home country is not relaxing. Maybe it's just money?
It’s not up to them. It’s up to people on random bulletin boards.
Hah
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
Probably I'm alone in PB. on that
That would be like wanting to join Italy provided we could still have regulations applied properly regardless of who you happen to know, could eat pasta where the sauce has been added after cooking rather than during, could order a cappuccino after 11 am, didn’t have to dress up in our best clothes and wander about the town every evening, and didn’t have to listen to what our mother is telling us.
Yes, we could do that, but we wouldn’t be Italian.
Yeah, but we'd be allowed to live in Italy
I don't want to be American but I would love the freedom to travel and live anywhere in the USA, the American protection of Free Speech, and the dynamism of the American economy
And a nutter like Trump might just be willing to offer it for the glory of adding a large new state to the USA (or four states?), and 70m people, and a mighty world city, and great universities, and some stunning countryside, and the English Premier League. And Wimbledon. And scones jam and cream. Etc
That would instantly make Trump one of the great American presidents, and he's not known for disliking praise and fame
Sad times, the British exceptionalist meme of the ex-colonies gagging to rejoin the mother country degenerating into humping Trump’s leg to become the 51st state. Oh Starmer, what have you done?!
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
Pretty certain. There are a few keywords that are give-aways to the very low-effort people, and there is an 'over-flowery' style. If you were just reading one CV you might not spot it, but when you have 30 with exactly the same patterns in it you begin to spot them.
Also the bargain-basement 'jazz up my CV' tools seem to just *love* adding nonsensical metrics and KPI's. "I improved the API response by 15% leading to an increase in user satisfaction of 37%" without any context. Just 'the API', 'user satisfaction' can be any plausible sounding thing. Again, if you were just reading one CV it might just roll past your eyes, but CV after CV after CV with the same lines.
ChatGPT has a very overly positive style, everything is lovely and wonderful, as a default to rewriting things. Claude on the other hand is overly verbose, ask it to turn bullet points into text and you end up with 24 pages of prose. You can of course be smarter about the prompts you use. I don't find the style tags that Claude have introduced very helpful.
It’s not up to them. It’s up to people on random bulletin boards.
Hah
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
Probably I'm alone in PB. on that
That would be like wanting to join Italy provided we could still have regulations applied properly regardless of who you happen to know, could eat pasta where the sauce has been added after cooking rather than during, could order a cappuccino after 11 am, didn’t have to dress up in our best clothes and wander about the town every evening, and didn’t have to listen to what our mother is telling us.
Yes, we could do that, but we wouldn’t be Italian.
Yeah, but we'd be allowed to live in Italy
I don't want to be American but I would love the freedom to travel and live anywhere in the USA, the American protection of Free Speech, and the dynamism of the American economy
And a nutter like Trump might just be willing to offer it for the glory of adding a large new state to the USA (or four states?), and 70m people, and a mighty world city, and great universities, and some stunning countryside, and the English Premier League. And Wimbledon. And scones jam and cream. Etc
That would instantly make Trump one of the great American presidents, and he's not known for disliking praise and fame
Sad times, the British exceptionalist meme of the ex-colonies gagging to rejoin the mother country degenerating into humping Trump’s leg to become the 51st state. Oh Starmer, what have you done?!
It could be a nice solution to your desire to escape the dreadful colonial yoke of England, which basically enslaves you by giving you more money than England and allowing you to vote on English laws whereas the English can't do the opposite. With my idea you could be the 52nd state, Wales 53 etc
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
Pretty certain. There are a few keywords that are give-aways to the very low-effort people, and there is an 'over-flowery' style. If you were just reading one CV you might not spot it, but when you have 30 with exactly the same patterns in it you begin to spot them.
Also the bargain-basement 'jazz up my CV' tools seem to just *love* adding nonsensical metrics and KPI's. "I improved the API response by 15% leading to an increase in user satisfaction of 37%" without any context. Just 'the API', 'user satisfaction' can be any plausible sounding thing. Again, if you were just reading one CV it might just roll past your eyes, but CV after CV after CV with the same lines.
ChatGPT has a very overly positive style, everything is lovely and wonderful, as a default to rewriting things. Claude on the other hand is overly verbose, ask it to turn bullet points into text and you end up with 24 pages of prose.
I'm finding the new-ish 'concise mode' on Claude quite good. Really tones down the verbose chatty 'voice' it adopts.
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
You could pick that up at the interview, surely? Tell us what a great team player you are by taking personal credit for everything your team accomplished in the last five years.
While it may be true that there is no I in team, it is also true that the team is not looking for a job offer!
For what it's worth, my work fell off a cliff this year, mostly down to AI.
Went from being excessively busy around Christmas 2023 to having very little luck finding anything so far after losing two long term clients in the first half of 2024.
And, if I can play my violin a moment, I don't have quite the mountain of wealth to just retire that some here do. On the plus side, I'm not destitute either but the situation isn't great and hasn't been for a while.
One of the reasons behind me starting the Undercutters podcast, as well as hopefully turning it into a side gig down the line, is that I can just crack on with it without AI eating it, and without having to beat 100-200+ other people applying for something.
I can't see how AI will do anything other than revolutionise some jobs. At the moment I'm between major projects so I've been looking into how it might be used in Project Management, and actually I've been disappointed as there doesn't yet seem to be much real benefit. All the examples of its use have been to generate templates of documents used in management (project plan, risk log) but if you work for any decent consultancy you already start from having all the templates you need. I've seen nothing yet which would make any difference to what I actually do for a living, which is mostly thinking and talking to people.
I'd be very interested to know if others in IT/Project Management have been able to get real benefit from AI yet?
(I would like an AI that could manage MS Project for me....)
I've been using it for a few IT/HR/Project things.
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
Thank you, your experience is useful to know. Triaging Help Desk tickets seems like a good application for AI.
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
The last post we advertised for I'd say 30-40% of the CV's were clearly written (or rewritten) by an AI tool. Even just a glance at them and you could spot them. Which makes me feel much less guilty about using an LLM to process them in return (there is a step that asks it to judge if the CV is likely LLM generated now too...)
Are you sure it was AI and not just morons writing the usual "I am a goal-centred team player" bilge?
Pretty certain. There are a few keywords that are give-aways to the very low-effort people, and there is an 'over-flowery' style. If you were just reading one CV you might not spot it, but when you have 30 with exactly the same patterns in it you begin to spot them.
Also the bargain-basement 'jazz up my CV' tools seem to just *love* adding nonsensical metrics and KPI's. "I improved the API response by 15% leading to an increase in user satisfaction of 37%" without any context. Just 'the API', 'user satisfaction' can be any plausible sounding thing. Again, if you were just reading one CV it might just roll past your eyes, but CV after CV after CV with the same lines.
ChatGPT has a very overly positive style, everything is lovely and wonderful, as a default to rewriting things. Claude on the other hand is overly verbose, ask it to turn bullet points into text and you end up with 24 pages of prose.
Comments
I think we are a little away from AI doing any of that.
I think one big question is how Copyright will apply, and where the limitations will be drawn. For the UK, where does consumption and summarising of others' material fall under Fair Dealing, given that quotes and citations are limited to what is necessary.
We've been through this before. I'd draw a parallel with our press having routinely copied others' content to within an inch of what they think they can get away with along the line of the edge of enforcible law.
I'd suggest that the relationship between the actor and their AI bit will reflect that between Principal and Agent.
Here's a piece about it on Techradar:
https://www.techradar.com/computing/artificial-intelligence/i-tried-google-s-new-one-click-ai-podcast-creator-and-now-i-don-t-know-what-s-real-anymore
Compute power will only grown over time.
We're already at the stage where a directed AI can perform specific cognitive tasks at a level well beyond most people. In that sense doing what the machine age did for manual labour.
We've not really worked out what that equivalent means for cognitive work - and we're only a few years into the process. And compute advances far more quickly than did the development of mechanical systems.
Even assuming no conceptual breakthroughs (which is daft, as they will happen), it will remake economies and leave many of us stranded.
And it potentially concentrates power in the hands of a few, orders of magnitude beyond what machines did.
And watch out for a revolution in Myanmar. It's not exactly stable at the moment.
Our AI-overlords will never take Xmas from us!!!
Still hoping to get to Myanmar this winter. I don’t mind revolutions. I’ve seen a couple
Something like my Dissertation could not be done by AI as the amount of original research I did would be beyond its capacity. AI could certainly summarise secondary sources far more quickly than I ever could, but the work requires a lot of study of unpublished archival material.
https://www.facebook.com/groups/1025657710790236/permalink/9090190431003550/?
She could become a translator, not least a she speaks English, Welsh, Italian, French and is passable in Japanese, but she seems to want to transition to law
I suspect TTRPGs might be more resilient than most areas to AI as it's about interaction with other people. I'm hoping the desire for authenticity in podcasts might help them hold up too.
https://x.com/yejinjgim/status/1871120204077715542
While former Defense Intelligence Commander Roh Sang-won is suspected of being involved in planning the martial law emergency on December 3, a memo suggesting "inducing North Korean attacks at the Northern Limit Line (NLL)" was found in his notebook...
..It was also discovered that the expression "shoot to kill" was written in former Commander Roh's notebook.
https://www.youtube.com/watch?v=0wrQLDoJ6cw
The job that disappeared was to a well known American magazine. It would astonish you if you knew the title. AI took the job, everyone accepts this
As for the “good” AI output, this isn’t a one off. In the last month I’ve seen AI produce this quality in multiple ways and in multiple subjects. It’s not just OK it is genuinely good with flashes of brilliance. The improvement is sensational and scary - and, as @SouthamObserver notes - is only going to improve
I find this, personally, deeply sad. This is my beloved profession, largely vanishing. It’s very hard to take for me as a veteran, I wholly understand why @Big_G_NorthWales doesn’t want to disillusion his grand-daughter. Who wants to convey this news to eager young talented people?
I barely know what to tell my own kids. They are both very bright and would normally be looking at cerebral, “cognitive” careers. God knows what they will do in reality
Maybe we will rebel and their jobs will be Fighting the Machines!
May I take this opportunity of wishing you and yours a very Happy Christmas
Indeed to all my fellow posters
A building story in our area is the number of care homes that are shutting, especially the smaller ones. This is mainly being driven by the inability to provide anything like the level of staffing, training, monitoring and form filling needed for the sort of money that local authorities can afford to pay. The care homes also have to fund the regulators that are putting them out of business.
One home that we approached about my Mother in law would only accept new residents who could prove they had 2 years worth of fees in the bank. They were not willing to accept the risk of a new resident who would only be producing the local authority rate after a relatively short period. Any home with a majority of authority paid residents is facing closure or prosecution. It is not a great choice.
It is also why we have the most expensive child care in the world making it very hard for many parents to find work that covers the cost.
What Kerry wasn't of course was an Obama or Reagan - a candidate with the political skills to bend a race their way by force of personality. But those are rare, generational figures. Of conventional politicians you could add Clinton (Trump rather sits in his own category) - but he did get lucky with third party candidates and his opponents. Otherwise it goes along with the fundamentals and we tend to malign candidates who lost as "bad" when they may have actually been middling or in some cases decent, but defeated by the headwinds.
Harris I think sits in that middling area - she didn't have the skills to dig her party out of the hole they'd dug by hanging on to Biden, but not a bad one who made obvious screw ups that affected the race. Plus one with a profile that was unhelpful in terms of being a Trump-killer. Basically, as with Biden in 2020, you needed to select the candidate with the most appeal to the Mid-west, as that's the entire ball game - and could've saved the Democrats even as the national picture shifted Trump's way.
But to say was a "bad candidate" probably lets the Democrats off the hook when it wasn't particularly anything she did that lost them the election - in fact she improved their position, just not by enough. When they should really be going back over the past 4 years and examining the decisions that put them in such a difficult place that Harris wasn't able to magic away with force of personality.
If we outsourced regulation to AI, I can see the AI deciding to exterminate the human race through rules banning procreation and then continuing to even more pointless regulations just to keep itself busy.
Maybe I'll write a play about it?
As a homage, I think I'd call the main AI character Sir Humphrey ...
Did Biden hang too long to give her a reasonable shot?
A longer run-in might have seen Trump win by an even bigger margin?
First - a job applicant scanner. Takes in a job description, some 'not on the description, but useful pointers to what we're after', and a candidates CV. Runs various focussed tasks on it - extracts key details, spot any red flags, reformat some of the info to make comparisons easier, outputs some details in the format our HR system wants, etc. Gives an overall summary with a 'call for interview? Y/N' along with it's reasoning. We've only been using it as a guide so far but it's *really* effective. A few people we've gone on to hire have been shown their own reports and have all agreed with the findings. So probably going to go into production in 2025. It can process about 100 CV's in a few minutes at a cost of about a quid.
Second - a 'user story generator'. Lets users of our in-house applications log in, ask for new features/bug-report/whatever about an app (it has the details and docs for each app to refer to as context), the LLM can ask some clarifying questions (or even say "the app already does that - click X then Y then Z'). Those user stories, the LLM's summary of the request, and the original conversation are all added as a new issue for the developers to then look over. So far it's working ok, needs a bit of refining to be properly useful. But certainly looks like it's going to cut down a lot of meetings/calls/overhead. The end users are also really liking it. The cost is so low it's effectively 'free'.
Third - auto-allocating helpdesk tickets. Very much a proof-of-concept (I only wrote it a few weeks ago). We give the LLM a list of our teams and what they do, and a list of the people in those teams and their specialities. When a new ticket comes in the LLM reads it, tries to figure out which team/people would be best suited and allocates the ticket to them. So far it's not actually allocating the ticket - it just adds a note saying who it thinks it should go to (and why). We'll monitor that to see how it compares to the people who do it by hand just now. I'd also like to give it the ability to search for similar tickets and use that information too - but we'll see how it goes.
Lastly - more an IT admin tool - our LLM system log scanner. We pipe our back-end system logs through the LLM which is given some specific guidelines about our tech stack. Then the LLM produces a report of 'interesting/odd/noteworthy' stuff that's happened. (We have specific tools for security/hardware faults with known messages outside of this). After it's highlighted the issue, gives an example of the log entry, which server(s) it's happening on etc, it offers some very brief technical hints as to what might be the underlying issue, how to investigate or fix it. It's really very useful. We're managing to clear up just a lot of little issues in minutes that otherwise might never have even been looked at.
https://x.com/Sports_Doctor2/status/1854071951381176355
Had he stepped back before the primaries, she'd either have been a far more credible candidate in the general, or someone diss would have won the nomination.
Number of times Nigel Farage appears as a panelist on Question Time in 2025
Moving away from AI itself to the economics of the future. You only have to crunch the maths to see what’s coming
Imagine a company making creative widgets. Let’s call them artburgers. It can employ a team of ten artburger designers each on £100 a day, each designing one art burger a day. Or it can get a droid to churn out those ten artburgers a days basically for free - and sack nine artburger designers and
keep one designer to polish up the ones made by the droid
Unfortunately for Artburger Inc the new mechanised artburgers aren’t as good as the man made ones. But then Artburger Inc makes the happy discovery that the consumers don’t care about the difference, indeed - they don’t even notice it. Only the pro human designers can spot the difference (and in the end that difference will disappear)
So Artburger Inc can now save £900 a day, £234,000 a year by sacking 90% of its designers. It will absolutely do that, not least coz its competitors will do the same
Trump says US owning Greenland ‘absolute necessity’
https://thehill.com/homenews/administration/5053319-trump-greenland-purchase/
Will different AI's have different feels'. Will the same AI have a different 'feel' every time it is used? Does the fact it is AI remove the human emotion and render it pointless?
I have lots of questions but no answers.
Fears Holyhead port could be 'closed until March'
https://www.dailypost.co.uk/news/north-wales-news/fears-holyhead-port-could-closed-30646009#ICID=Android_DailyPostNewsApp_AppShare
You can see this having an effect in almost every market. In the two year gap between their Ryzen 7000 and 9000 processors AMD managed to get an average performance uplift of just 5%. NVidia is resorting to software-based tricks like frame generation to boost performance of their GPUs at the cost of image quality. Their upcoming 5000-series graphics cards are rumoured to be using an 'AI' texture compression system, again trading image quality for performance.
At the moment LLM based AI simply uses too much hardware and consumes too much power, the solution to that is not likely to come from better hardware.
It's one thing helping the customer to understand what they want and how it can be achieved (or why it can't be achieved in the way they hoped), but when you get developers deciding that the customers don't know what they're talking about in their own business it's a big problem.
Good afternoon, everybody.
Has he found his North-West passage yet?
Chump's trying to behave like a late-Victorian, when the USA just started, or manufactured, a war (ish) if they wanted anything near the N/S American continents.. He was born in 1946, at Junior/Senior School (whatever the Yanks call it) in the 1950s when it was Communism and McCarthy.
He's imposing what he thinks he knows, rather than bothering to find out about the things he doesn't know.
It will hit a problem imo when he finds out that the USA is not as hegemonic as it arguably was just after WW2.
See also:
Farage, Nigel
Journalists, Log Rolling
Murdoch, Rupert
Musk, Elon
News, GB
Non-Dom, Lord
Rees Mogg, Jacob
Spectator, Contributors to
I had heard about AI being used to scan CVs recently from people seeking new roles and how it was making it harder for them (the flip side, perhaps, of being better for employers).
I treat its replies as hints rather than deterministic replies: most of them are "good enough", but many havce errors and some have fatal errors. I don't have a ratio for you I'm afraid.
EHCP's have been a failure.
We need more inclusion and an end to the division of SEN/non-SEN.
Increasingly just a form of exclusion and exclusivity.
https://www.theguardian.com/uk-news/2024/dec/23/send-children-special-educational-needs-disabilities-anne-longfield
It's not free. Am not interested.
I saw this appear just the other week https://gitingest.com/ and am considering using it to give the LLM even more context, so it can add much better breakdowns to the developers. Luckily our developers are pretty good at understanding user requests - and if there's a 'whuuu?' will go back to the user to clarify.
I don't see any particular road-blocks in making it more fleshed out in other directions though. It's effectively free at our scale. A whole to-and-fro with the user and LLM is a fraction of a fraction of a penny.
Joggers
disabled people
deer
The last one has been altered to read "old deers".
If Trump offered the UK American statehood, and we were allowed to swerve American gun law, avoid the American healthcare system, and find some compromise over the monarchy where we get to keep it but they don't have to adopt it, I would absolutely say Yes
Probably I'm alone in PB. on that
It's really quite impressive. I don't use most of it's features - but the raw editing experience is quite something. They've really put a lot of effort into making it understand what you're trying to do - rather than just guess at a likely autocomplete a-la copilot.
A few weeks ago I was contemplating a header arguing why it will take 50-100 years for Syria to get back to anything other than an uneasy peace, and thinking about different communities / groups understanding each other, and how long that takes just as a starter in a peace process.
One analogy I was thinking about for understanding and misunderstanding was the use of "unicorn" in the King James Bible Isaiah 34:14, and how the translation differs according to the values of the translator, the target community / audience, and the type of translation intended (study, for non-native speakers, to read aloud, literal, aimed at concepts or underlying meaning not precise language etc.)
The nearest term is probably something around wild ox, but terms used in English translations include the strange "unicorn" (which then leaks from the KJV into for example The Book of Mormon, as the writer rather fetishises the King James Bible), "rhinoceros", "wild buffalo" (USA, I assume) and others. That is then overlaid with a process by some modern-speculators trying to read the word-choice as an indication of classical mythology as a source for the original documents to justify their New Age beliefs.
Greg Wallace got a look in, because the same KJV also includes "satyrs", half-clothed man-goats.
Understanding these complications is one challenge for AI. Quality checking it may be more challenging.
There is a lovely table, which I cannot find again, of the different words. My photo quota today is how one place illustrates "rhinoceros".
Codeium is another one that people recommend. I think you know the market might be getting to frothy when a VS Code fork + some filtering of existing LLM gives you so much f##k up start up money you can make the ad for your IDE as a spoof on Silicon Valley tv show and hire one of the actors from that show.
What are the chances of Labour going sub-20 in a poll next year?
Looking at the already grim economic news allied with their total inteptitude, their complete lack of ideas, the arrival of Trump, the slowdown in Europe, the horror that is Starmer and Reeves, and the rise of Farage stealing Labour voters, I reckon there's a 40-50% chance
And once you go that low it gets pretty hard to come back
A good spectator sport.
Yes, we could do that, but we wouldn’t be Italian.
I don't want to be American but I would love the freedom to travel and live anywhere in the USA, the American protection of Free Speech, and the dynamism of the American economy
And a nutter like Trump might just be willing to offer it for the glory of adding a large new state to the USA (or four states?), and 70m people, and a mighty world city, and great universities, and some stunning countryside, and the English Premier League. And Wimbledon. And scones jam and cream. Etc
That would instantly make Trump one of the great American presidents, and he's not known for disliking praise and fame
I sometimes become a little too intense in my constant striving for perfection.
https://www.youtube.com/watch?v=G3zirE_r5l4
Also the bargain-basement 'jazz up my CV' tools seem to just *love* adding nonsensical metrics and KPI's. "I improved the API response by 15% leading to an increase in user satisfaction of 37%" without any context. Just 'the API', 'user satisfaction' can be any plausible sounding thing. Again, if you were just reading one CV it might just roll past your eyes, but CV after CV after CV with the same lines.