Meeks is very readable and makes some excellent points. But the conclusion I draw from his analysis is that every party is doing badly.
So the question ultimately comes down to a political one. Will the viewers decide Labour have failed and need to be replaced? And, if they do, will they trust Reform to be the instrument to kick Labour out?
If the answers to those questions are "yes" and "no" then the Tories have a decent chance of winning back votes and seats. And if they aren't, then at least the Tories under Badenoch have a simple message around which to fight a rearguard action: "sound money for lower taxes."
I did think this paragraph was astute, though.
At the last election the Conservatives won 121 seats. This is by far their worst result since the Great Reform Act of 1832. Despite that, it seems not to be understood just how bad a result it was.
Realistically the Tories cannot hope to do well at the next election, barring a miracle (or, well, multiple disasters for their opponents). So the most that Badenoch can be expected to achieve is to reduce the damage.
This does leave us with a very large landing zone for a hung Parliament.
What exactly from the previous 14 years of government would persuade anyone that "sound money for lower taxes" is the Tory Party USP?
"Shortterm austerity measures resulting in runaway cost escalations and increasing tax take" would be the succinct summary of their 14 years.
Burnham in first test of his resistance to rightwing media pressure with decision on ERS response to Conservative caused prisons crisis.
"Al Carns refuses to join Burnham’s Government over defence spending Former Armed Forces minister resigned last month because he did not believe military had been offered enough money"
It's another Keir Starmer. Man with unarguably stellar record in another career having rings run round him in politics, because he isn't a politician. It's a craft. Which takes time and learning, particularly from mistakes. We used to know that. Now it seems to be vault to the top. Don't cut it, unsurprisingly. Career over. Having failed to win the Labour leadership twice is probably extremely good experience for your third go.
Being rejected twice, so not doing the job, is good experience for doing the job? LOL!!
Meeks is very readable and makes some excellent points. But the conclusion I draw from his analysis is that every party is doing badly.
So the question ultimately comes down to a political one. Will the viewers decide Labour have failed and need to be replaced? And, if they do, will they trust Reform to be the instrument to kick Labour out?
If the answers to those questions are "yes" and "no" then the Tories have a decent chance of winning back votes and seats. And if they aren't, then at least the Tories under Badenoch have a simple message around which to fight a rearguard action: "sound money for lower taxes."
I did think this paragraph was astute, though.
At the last election the Conservatives won 121 seats. This is by far their worst result since the Great Reform Act of 1832. Despite that, it seems not to be understood just how bad a result it was.
Realistically the Tories cannot hope to do well at the next election, barring a miracle (or, well, multiple disasters for their opponents). So the most that Badenoch can be expected to achieve is to reduce the damage.
This does leave us with a very large landing zone for a hung Parliament.
What exactly from the previous 14 years of government would persuade anyone that "sound money for lower taxes" is the Tory Party USP?
They cut inheritance tax and the top rate of income tax and the lowest earners out of tax and the Coalition reduced spending
And haven't raised thresholds for years. Sorry. It won't wash.
"Al Carns refuses to join Burnham’s Government over defence spending Former Armed Forces minister resigned last month because he did not believe military had been offered enough money"
It's another Keir Starmer. Man with unarguably stellar record in another career having rings run round him in politics, because he isn't a politician. It's a craft. Which takes time and learning, particularly from mistakes. We used to know that. Now it seems to be vault to the top. Don't cut it, unsurprisingly. Career over. Having failed to win the Labour leadership twice is probably extremely good experience for your third go.
Being rejected twice, so not doing the job, is good experience for doing the job? LOL!!
No. It's the only experience for getting the job. Applying twice. Being rejected. Hasn't fucked the application up a third time. Has actually executed a takeover almost no one thought possible. Doesn't make him better at the job. But it sure as hell made him better at getting it.
File on 4 Investigates has also learned from several sources that the Crown Prosecution Service did not disclose the obstetrics records to the defence.
The CPS reasoned that because the prosecution did not use them to form its case against Letby, they were not relevant.
In other words, the CPS didn't include obstetrics records as part of the prosecution case because they were 'unhelpful', and therefore withheld them from the defence because they were 'irrelevant'.
Thats a pretty feeble excuse by Letby's lawyers. Obviously there would be obstetric notes for each mother and the Defence could have requested them if they wanted them.
They did. That's not good.
The BBC has also seen evidence the defence did request the maternity notes but they were not released.
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It was 58.2 million in 1996 and it's approx 69.5 million today, so you could just about argue it's (one) tens of millions.
Of course, not all is immigration but a lot is.
I suspect our population would have declined without immigration. Ovet 10m seems right to me. (I suspect tge true population is rather more than 69.5m today).
“Tens of millions” has to mean at least 20 million, or close to 20 million. ~10 million does not count. Yusuf lied.
No it does not. It is the interval of 10,000,000 < x < 100,000,000
1.01 x 10⁷ is in the tens of millions, just as the decade the 2010s was 2010 to 2019.
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It's not necessarily hogwash. The *net* population increase is about 12 million, but there's a lot of churn in that. It's quite possible from the avaliable numbers that over 20 million individuals have entered in the last 30 years, even if some of them (and a chunk of the existing British population) have also left.
Weird to call it hogwash for using an order of magnitude, its quite common to refer to large numbers by their order of magnitude.
Tens of millions covers anything between 10 million and less than 100 million anyway, it does not begin at 20 million.
Given everything we know, its almost certainly the right order of magnitude.
2-10 million is millions. 20-100 is tens of millions. 200,000-1 million is hundreds of thousands.
1 million < x < 10 million is millions
1 million is exactly 1,000,000 1,000,001 is in the millions.
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It's not necessarily hogwash. The *net* population increase is about 12 million, but there's a lot of churn in that. It's quite possible from the avaliable numbers that over 20 million individuals have entered in the last 30 years, even if some of them (and a chunk of the existing British population) have also left.
Weird to call it hogwash for using an order of magnitude, its quite common to refer to large numbers by their order of magnitude.
Tens of millions covers anything between 10 million and less than 100 million anyway, it does not begin at 20 million.
Given everything we know, its almost certainly the right order of magnitude.
2-10 million is millions. 20-100 is tens of millions. 200,000-1 million is hundreds of thousands.
1 million < x < 10 million is millions
1 million is exactly 1,000,000 1,000,000 is 1.000001 millions
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It was 58.2 million in 1996 and it's approx 69.5 million today, so you could just about argue it's (one) tens of millions.
Of course, not all is immigration but a lot is.
I suspect our population would have declined without immigration. Ovet 10m seems right to me. (I suspect tge true population is rather more than 69.5m today).
“Tens of millions” has to mean at least 20 million, or close to 20 million. ~10 million does not count. Yusuf lied.
Yusuf probably isn't lying. With the huge inflows and outflows of people, it's quite probable that over 20 million immigrants have come here at some point in the last 30 years, but the outflow of emigrants has offset that so the net population increase is *only* 12 million (which is of course at least 12 million more than it should have been).
Ancidata on this - I worked with quite a lot of Poles doing fairly low skilled factory work between 2008-12. Not all, but quite a lot of them are back in Poland now.
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It was 58.2 million in 1996 and it's approx 69.5 million today, so you could just about argue it's (one) tens of millions.
Of course, not all is immigration but a lot is.
I suspect our population would have declined without immigration. Ovet 10m seems right to me. (I suspect tge true population is rather more than 69.5m today).
“Tens of millions” has to mean at least 20 million, or close to 20 million. ~10 million does not count. Yusuf lied.
Yusuf probably isn't lying. With the huge inflows and outflows of people, it's quite probable that over 20 million immigrants have come here at some point in the last 30 years, but the outflow of emigrants has offset that so the net population increase is *only* 12 million (which is of course at least 12 million more than it should have been).
Ancidata on this - I worked with quite a lot of Poles doing fairly low skilled factory work between 2008-12. Not all, but quite a lot of them are back in Poland now.
If you put "total gross number of inward migrants and asylum seekers to the UK since 1996" into Google, its AI gives you "There is no single official, aggregated cumulative total for gross inward long-term migration and asylum seekers spanning 1996 to the present. However, based on data from the Office for National Statistics and the House of Commons Library, cumulative inward long-term migration from 1997 through 2025 exceeds 15 million people, with over 2.5 million asylum claims lodged since 1996."
So yes, Yusuf is correct. Annoying, self-pitying, aggressive to others, whiny to himself, incapable of forming an argument without complaining, somebody who should not be let anywhere near power, yes, but in this he is correct (assuming the AI is right).
Also note this is gross inward movement, not net. After taking gross outward movement into account, the net inward movement will be smaller. But - to belabour the point - we're back to arguing about the side of the bus again.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
Tens of millions? oh FFS. At least get the basic numbers right. How big was the number in last thirty years? Thirty million imported? Forty million? Eight five?
Utter hogwash on stilts
It was 58.2 million in 1996 and it's approx 69.5 million today, so you could just about argue it's (one) tens of millions.
Of course, not all is immigration but a lot is.
I suspect our population would have declined without immigration. Ovet 10m seems right to me. (I suspect tge true population is rather more than 69.5m today).
“Tens of millions” has to mean at least 20 million, or close to 20 million. ~10 million does not count. Yusuf lied.
Yusuf probably isn't lying. With the huge inflows and outflows of people, it's quite probable that over 20 million immigrants have come here at some point in the last 30 years, but the outflow of emigrants has offset that so the net population increase is *only* 12 million (which is of course at least 12 million more than it should have been).
Ancidata on this - I worked with quite a lot of Poles doing fairly low skilled factory work between 2008-12. Not all, but quite a lot of them are back in Poland now.
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Here is a gift link to bypass the paywall:-
Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
In addition to books, there seems to be a lot of video history/documentary content that I'm guessing has been scripted by AI by stitching together secondary and tertiary sources. It's got that same overblown, creative writing 101 feel, not helped by synthesised narration (which itself is getting better at not tripping over numbers).
The pressure was intense, the stakes were high, and failure was not an option.
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Doesn't he do the Speccie any more? Does the Telegraph pay more?
Why not both?
iirc he was a Telegraph regular until they laid off a number of columnists. He still contributes occasionally on what I guess is a freelance basis. That left the Spectator and the Gazette.
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
They only came in as a temporary measure after the terrorist attacks on various airports. Andy B, come on, give the greedy bastard airport a kicking and force them to go back to nought or £1.
They only came in as a temporary measure after the terrorist attacks on various airports. Andy B, come on, give the greedy bastard airport a kicking and force them to go back to nought or £1. Would be really popular.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
I can't even conceptualise it. I know how to do email: there's a screen, I open an account, I send and recieve emails. I know how to write a R program: I open RStudio, I write a program in R, I run it. Similarly for Stata and SAS. I used to know how to run Python: open PyCharm, write Python code, I run it. But I asked rcs1000 about AI and I got this:
"...Get yourself a $5 VPS. Put a web based SSH server on it. Login. Install Claude Code or Codex or whatever. The cheapest way to get started is probably OpenCode + one of the really cheap Chinese models on the backend..."
Which is detailed and probably intended to be helpful, but it had me dribbling like somebody who's had a stroke. I don't know how to do any of that.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
I can't even conceptualise it. I know how to do email: there's a screen, I open an account, I send and recieve emails. I know how to write a R program: I open RStudio, I write a program in R, I run it. Similarly for Stata and SAS. I used to know how to run Python: open PyCharm, write Python code, I run it. But I asked rcs1000 about AI and I got this:
"...Get yourself a $5 VPS. Put a web based SSH server on it. Login. Install Claude Code or Codex or whatever. The cheapest way to get started is probably OpenCode + one of the really cheap Chinese models on the backend..."
Which is detailed and probably intended to be helpful, but it had me dribbling like somebody who's had a stroke. I don't know how to do any of that.
Dell Pro Max with GB10 is another machine touted for AI.
There will be videos on YouTube offering guidance, although I cannot now find the one I was looking for.
I think FrancisUrquhart has posted here about using AI at work. ETA and now I see he has just posted.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
The SeanT articles about the AI books flooding Amazong...Dead Internet theory in action, will LLMs be accidentally trained on all these fake books with all the fake info?
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Here is a gift link to bypass the paywall:-
Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
You want to have the installed version, this is where the real power is. Get yourself a mac mini (or use a VM on your normal PC), slap on Claude Code, and just experiment. Don't do it on a main pc or an un-VMs sandbox, as although it is pretty good at asking permission before doing anything too crazy, its better to have a sandpit where you can cause mayhem without worrying that just deleted all your photos of your dead dog.
They only came in as a temporary measure after the terrorist attacks on various airports. Andy B, come on, give the greedy bastard airport a kicking and force them to go back to nought or £1. Would be really popular.
I'd be more worried about recent advice to turn up several hours early because the new border checks are broken.
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
From Starmer's early days as Prime Minister, I was forecasting that he would retire early like Wilson. He often made slips, and in recent speeches seems to be reading if not word-by-word, then certainly with frequent glances at his notes. Aside from that, I see Starmer as a lawyer, not a politician; as a technocrat who came into office believing the civil service already had the answers but was prevented from carrying them out by incompetent and malicious Conservatives. I do not really see the comparisons with Rishi and John Major.
What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Here is a gift link to bypass the paywall:-
Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
I just read it and it's quite sad (as in touching, not pathetic). He made some good points about there being so many books the concept of reading becomes silly.
But the fact that they mentioned "the Edgerton book" meant that they and I had some common conceptual ground which in turn meant that I could theoretically join the discussion if invited. What do we - what do I - do when we don't have books in common?
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
You want to have the installed version, this is where the real power is. Get yourself a mac mini (or use a VM on your normal PC), slap on Claude Code, and just experiment. Don't do it on a main pc or an un-VMs sandbox, as although it is pretty good at asking permission before doing anything too crazy, its better to have a sandpit where you can cause mayhem without worrying that just deleted all your photos of your dead dog.
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Here is a gift link to bypass the paywall:-
Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
I just read it and it's quite sad (as in touching, not pathetic). He made some good points about there being so many books the concept of reading becomes silly.
But the fact that they mentioned "the Edgerton book" meant that they and I had some common conceptual ground which in turn meant that I could theoretically join the discussion if invited. What do we - what do I - do when we don't have books in common?
And the books problem – if I might be permitted a small rant on education in ydoethur's absence – is made worse by, aiui, school English courses increasingly dealing with fragments of texts; and history no longer being a chronological tale from cavemen to 30 years ago, but now in a few modules jumping all over the timeline, often from abroad rather than here, and being more concerned with teaching pupils to weigh evidence from different sources than with teaching what happened. (Of course, weighing different sources is probably a useful life skill.)
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
From Starmer's early days as Prime Minister, I was forecasting that he would retire early like Wilson. He often made slips, and in recent speeches seems to be reading if not word-by-word, then certainly with frequent glances at his notes. Aside from that, I see Starmer as a lawyer, not a politician; as a technocrat who came into office believing the civil service already had the answers but was prevented from carrying them out by incompetent and malicious Conservatives. I do not really see the comparisons with Rishi and John Major.
What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
Also, I think their thesis about Starmer's majority being due to the implosion of the Conservatives as much as Starmer's changes to Labour, misses the point it is the NOTA parties that directly gained, with Labour's landslide coming from fewer votes than Corbyn's Labour attracted in the two previous general elections. (The LibDems, ironically, gained what for them is a landslide but from fewer votes than Reform, something they are strangely quiet about.)
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
You want to have the installed version, this is where the real power is. Get yourself a mac mini (or use a VM on your normal PC), slap on Claude Code, and just experiment. Don't do it on a main pc or an un-VMs sandbox, as although it is pretty good at asking permission before doing anything too crazy, its better to have a sandpit where you can cause mayhem without worrying that just deleted all your photos of your dead dog.
You run the relevant command, and it guides you through the setup. It is basically 2 steps. And then once its installed and your account verify, you can just type claude in any directory etc and then you will be in the world of Claude Code.
Its works far better with Linux / Mac than WIndows, because of the much more powerful command line terminals.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
You want to have the installed version, this is where the real power is. Get yourself a mac mini (or use a VM on your normal PC), slap on Claude Code, and just experiment. Don't do it on a main pc or an un-VMs sandbox, as although it is pretty good at asking permission before doing anything too crazy, its better to have a sandpit where you can cause mayhem without worrying that just deleted all your photos of your dead dog.
You run the relevant command, and it guides you through the setup. It is basically 2 steps. And then once its installed and your account verify, you can just type claude in any directory etc and then you will be in the world of Claude Code.
Very interesting. I'll have to look up what other maths problems have been solved by AI.
I've currently got it transcribing and cataloguing my radio drama / audiobook collection that goes back almost 30 years of mp3's and is full of 1950s+ recordings (a lot of them home-recorded on cassettes). After a few iterations it's now able to use a very efficient transcription model locally, take into the time of day for how much system load vs time we can trade off, all sorts of fanciness. All so I can ask a local DB "I think I have an old radio play with Dinsdale Landen where he plays a spy who....".
"We spent an entire month generating concept images for the motorcycle and car models, the tunnel and bridge, the surrounding mountains and even the ice floating on the river," Ryu said.
While the team delivered fewer than 300 final shots, they generated more than 2,000 during the production process. Ryu stressed that human expertise remains crucial.
"No matter how realistic the images look, viewers notice when the flow feels wrong," he said. "We approached it as if we were shooting live action, with particular attention to constructing and directing each shot using the visual grammar of cinema."
While the industry remains cautious, Ryu said an increasing number of creators are exploring AI video following the high-profile experiment with "Agent Kim Reactivated.".
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
The narrative of Stamer being super component technocrat that is a decent family man only in it for public service, the mask has slipped somewhat. Friendly journalists will always try to rewrite the narrative, they tried to do it after Brown. The reality is he got found out, he couldn't do the job, wasn't quite is only in it for the public service as he made out, was very good at claiming he never saw nothing, his card badly tarnished by Mandy scandal, while was pretty ruthless in sacking people and screwing over his own MPs.
That all been said, he ain't in the same universe of lying, cheating, only in it for themselves as Boris. Starmer was a level 1 to Boris level 100. And wasn't anywhere near as bad as Blair or Brown in terms of lying, dark arts, etc.
Interesting question for Burnham-Streetiing to address - will they allow Trump to continue using Fairford for the B52s to go on their bombing runs against Iran? Can anyone still hold the line that they are aren't being used for punishment beatings of the Tehran regime?
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
One of Mrs Foxy's cousins did a PhD in AI in the financial services industry at Southampton Uni, about 10 years ago, then went to work in Switzerland before moving back to London. He is very well paid and seems to be very valued by his bank.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
One of Mrs Foxy's cousins did a PhD in AI in the financial services industry at Southampton Uni, about 10 years ago, then went to work in Switzerland before moving back to London. He is very well paid and seems to be very valued by his bank.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
A PhD is totally different from a taught Masters. Southampton is quite a good research university so doesn't surprise me somebody getting a PhD from there ends up in a decent job. But its also worth noting, the world is very very different now. LLMs have completely disrupted academic research, even in the past 6 months, the world of doing research has completely shifted. Even in the past few weeks, the new ChatGPT model is now for the first time actually producing novel maths proofs* that academics are actually publishing and been peer reviewed as novel.
* there were claims upto a year ago of this, but were found that papers already existed but just unknown to those making the claims.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
One of Mrs Foxy's cousins did a PhD in AI in the financial services industry at Southampton Uni, about 10 years ago, then went to work in Switzerland before moving back to London. He is very well paid and seems to be very valued by his bank.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
A PhD is totally different from a taught Masters. Southampton is quite a good research university so doesn't surprise me somebody getting a PhD from there ends up in a decent job. But its also worth noting, the world is very very different now. LLMs have completely disrupted academic research, even in the past 6 months, the world of doing research has completely shifted.
Yes certainly different. Just saying that their department does have some expertise in the area.
The pronoun pattern hidden in Andy Burnham's first PM speech
Andy Burnham just delivered his first speech as Prime Minister. There is one pattern hidden in it. Once you see it, you cannot unsee it.
Watch the pronouns.
Every promise that can visibly fail starts with 'I.' 'I will bring forward a plan.' 'I can give people breathing space.' 'I will set out measures tomorrow.' 'I will issue my first instruction.'
Then the moment the goal gets bigger than one person, he flips to 'we.' 'We will change politics.' 'We will take power out of here.' 'We will build more council homes.'
Then the sentence the whole speech has been building towards: 'I will give this my all, and I ask you all to pull with me.' I takes the responsibility. We does the work. That line is the leadership contract.
Most leaders get this backwards. They say 'we' when things go wrong and 'I' when things go right. Burnham flips it. That is how you ask people to trust you before you have earned it. https://www.youtube.com/shorts/gt2Ofzye4WM
The pronoun pattern hidden in Andy Burnham's first PM speech
Andy Burnham just delivered his first speech as Prime Minister. There is one pattern hidden in it. Once you see it, you cannot unsee it.
Watch the pronouns.
Every promise that can visibly fail starts with 'I.' 'I will bring forward a plan.' 'I can give people breathing space.' 'I will set out measures tomorrow.' 'I will issue my first instruction.'
Then the moment the goal gets bigger than one person, he flips to 'we.' 'We will change politics.' 'We will take power out of here.' 'We will build more council homes.'
Then the sentence the whole speech has been building towards: 'I will give this my all, and I ask you all to pull with me.' I takes the responsibility. We does the work. That line is the leadership contract.
Most leaders get this backwards. They say 'we' when things go wrong and 'I' when things go right. Burnham flips it. That is how you ask people to trust you before you have earned it. https://www.youtube.com/shorts/gt2Ofzye4WM
He is a really good communicator. EQ is more important than IQ in politics. Burnham seems to have both.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
The pronoun pattern hidden in Andy Burnham's first PM speech
Andy Burnham just delivered his first speech as Prime Minister. There is one pattern hidden in it. Once you see it, you cannot unsee it.
Watch the pronouns.
Every promise that can visibly fail starts with 'I.' 'I will bring forward a plan.' 'I can give people breathing space.' 'I will set out measures tomorrow.' 'I will issue my first instruction.'
Then the moment the goal gets bigger than one person, he flips to 'we.' 'We will change politics.' 'We will take power out of here.' 'We will build more council homes.'
Then the sentence the whole speech has been building towards: 'I will give this my all, and I ask you all to pull with me.' I takes the responsibility. We does the work. That line is the leadership contract.
Most leaders get this backwards. They say 'we' when things go wrong and 'I' when things go right. Burnham flips it. That is how you ask people to trust you before you have earned it. https://www.youtube.com/shorts/gt2Ofzye4WM
He is a really good communicator. EQ is more important than IQ in politics. Burnham seems to have both.
Communicate all you like. You can't polish a turd.
If it's the sort of help that Britain got, it basically means KSA paying a great deal of money for nukes that depend on the USA and that they can't really use. That's safer for the USA than not helping them.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
Given the relatively small cost of the policy one wonders why Starmer and Reeves didn’t do this ? . Burnham clearly wants to make that positive first impression and get the public onside . It’s certainly good politics .
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
Given the relatively small cost of the policy one wonders why Starmer and Reeves didn’t do this ? . Burnham clearly wants to make that positive first impression and get the public onside . It’s certainly good politics .
Because they had just done the opposite and whacked up business rates which particularly hit pubs in nice old buildings. So they would have had to do another U-turn.
I think they didn't quite realise this. They were ending the big COVID discount and I don't think they appreciated that for lots of hospitality (already getting whacked by higher minimum wage and NI rises) they were plenty of case it lead to crazy rises in their business rates.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
This is a pretty good and interesting article imo.
Here is a gift link to bypass the paywall:-
Sean Thomas The author, the algorithm and the end of literary struggle AI-generated texts flooding Amazon should scare anyone who cares about the written word https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
The main consequence of that article will be to increase still further the number of folk doing exactly what he describes.
It’s bizarre that he has this weird idea that travel writing will escape the oncoming Armageddon for writers. Travel is an area where there’s so much information available online that AIs would find creating travel articles easy peasy.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
Yes. Claude weights have leaked to China in the past, but this particular model it appears they did some mass distillation which you can't really do via simple just firing up the web interface and asking questions.
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
1,460 soldiers and 81 artillery/MLRS not reporting for duty in Ukraine today.
Interestingly, also 28 tanks and 12 armoured fighting vehicles - biggest totals in a long while. Looks like the the generals were told to make a breakthrough. Ukraine was waiting.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
btw within Posix you can choose what variety of Claude meets your need/budget. Opus is available.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
btw within Posix you can choose what variety of Claude meets your need/budget. Opus is available.
Why would you do that, rather than just use Claude Code and use an MCP. All these companies trying to charge £20-30pm for each app, doesn't make a lot of sense when you can just install Claude Code and either via Cowork or MCP control the apps.
If it's the sort of help that Britain got, it basically means KSA paying a great deal of money for nukes that depend on the USA and that they can't really use. That's safer for the USA than not helping them.
Either way, it's the U.S. giving them the green light to enrich uranium, which effectively is permission to build their own nuclear weapons.
The lack of attached conditions (in contrast to the deal being negotiated by the last administration) doesn't help.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
btw within Posix you can choose what variety of Claude meets your need/budget. Opus is available.
Why would you do that, rather than just use Claude Code and use an MCP. All these companies trying to charge £20-30pm for each app, doesn't make a lot of sense when you can just install Claude Code and either via Cowork or MCP control the apps.
Because I don't have the nouse in that particular area, nor the time to work it up when there are other things to do
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
Yes. Claude weights have leaked to China in the past, but this particular model it appears they did some mass distillation which you can't really do via simple just firing up the web interface and asking questions.
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
I would add ByteDance for image and video generation. SeedDance 2 and 2.5 is ahead of Google / OpenAI video models.
Trump got discovery orderd against him for all his financial records, in his $10 billion claim against the BBC. The Beeb have sent out over 100 subpoenas covering family members and associates.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
btw within Posix you can choose what variety of Claude meets your need/budget. Opus is available.
Why would you do that, rather than just use Claude Code and use an MCP. All these companies trying to charge £20-30pm for each app, doesn't make a lot of sense when you can just install Claude Code and either via Cowork or MCP control the apps.
Because I don't have the nouse in that particular area, nor the time to work it up when there are other things to do
It is really easy. Claude Code will basically do all the setting up for you.
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
From Starmer's early days as Prime Minister, I was forecasting that he would retire early like Wilson. He often made slips, and in recent speeches seems to be reading if not word-by-word, then certainly with frequent glances at his notes. Aside from that, I see Starmer as a lawyer, not a politician; as a technocrat who came into office believing the civil service already had the answers but was prevented from carrying them out by incompetent and malicious Conservatives. I do not really see the comparisons with Rishi and John Major. What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
Radical in this sense meaning "exactly the same as all Labour chancellors since at least the 80s, pretending to be fiscally responsible whilst borrowing and spending money like a drunken sailor".
Then in January 2026, the government announced pubs and music venues in England would be given a 15% discount on their business rates bills from April and would not see increases for two years.
Is Andy's another 20% ontop, or simply going from 15% to 20%?
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
If it's the sort of help that Britain got, it basically means KSA paying a great deal of money for nukes that depend on the USA and that they can't really use. That's safer for the USA than not helping them.
Either way, it's the U.S. giving them the green light to enrich uranium, which effectively is permission to build their own nuclear weapons.
The lack of attached conditions (in contrast to the deal being negotiated by the last administration) doesn't help.
It'll give some future president somewhere extra to take military action against; finding places to use up shedloads of military ordnance every decade or so is essential to feed the US's military-industrial complex with new orders.
Then in January 2026, the government announced pubs and music venues in England would be given a 15% discount on their business rates bills from April and would not see increases for two years.
Is Andy's another 20% ontop, or simply going from 15% to 20%?
January 2026 surely meant April 2026.
We are now in July so Burnham is talking about April 2027 so it's an extra discount.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
One of Mrs Foxy's cousins did a PhD in AI in the financial services industry at Southampton Uni, about 10 years ago, then went to work in Switzerland before moving back to London. He is very well paid and seems to be very valued by his bank.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
A PhD is totally different from a taught Masters. Southampton is quite a good research university so doesn't surprise me somebody getting a PhD from there ends up in a decent job. But its also worth noting, the world is very very different now. LLMs have completely disrupted academic research, even in the past 6 months, the world of doing research has completely shifted. Even in the past few weeks, the new ChatGPT model is now for the first time actually producing novel maths proofs* that academics are actually publishing and been peer reviewed as novel.
* there were claims upto a year ago of this, but were found that papers already existed but just unknown to those making the claims.
LLMs have not completely disrupted academic research. LLMs are proving interesting tools in one area of maths research and helpful tools at speeding up searching literature and writing code moe generally.
Then in January 2026, the government announced pubs and music venues in England would be given a 15% discount on their business rates bills from April and would not see increases for two years.
Is Andy's another 20% ontop, or simply going from 15% to 20%?
January 2026 surely meant April 2026.
We are now in July so Burnham is talking about April 2027 so it's an extra discount.
When dealing with politicians, I always check the number of fingers I am left with after shaking their hands....
A betting account in the name of George Cottrell, the convicted fraudster who bankrolled Nigel Farage, received almost $9mn in cryptocurrency from unidentified sources while he was providing gifts and benefits to the Reform UK leader.
In October 2024, the account on Polymarket registered in the name “GCottrell93” received about $8.8mn from two separate wallets. The funds were deposited from two crypto wallets and then used to bet on Donald Trump winning the US presidential election.
Shortly after the election result, which generated $4.46mn in profit, almost all of the money in the account was emptied into two different wallets. It is unclear who provided the money used by the account and who ultimately received the winnings.
A betting account in the name of George Cottrell, the convicted fraudster who bankrolled Nigel Farage, received almost $9mn in cryptocurrency from unidentified sources while he was providing gifts and benefits to the Reform UK leader.
In October 2024, the account on Polymarket registered in the name “GCottrell93” received about $8.8mn from two separate wallets. The funds were deposited from two crypto wallets and then used to bet on Donald Trump winning the US presidential election.
Shortly after the election result, which generated $4.46mn in profit, almost all of the money in the account was emptied into two different wallets. It is unclear who provided the money used by the account and who ultimately received the winnings.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
From Starmer's early days as Prime Minister, I was forecasting that he would retire early like Wilson. He often made slips, and in recent speeches seems to be reading if not word-by-word, then certainly with frequent glances at his notes. Aside from that, I see Starmer as a lawyer, not a politician; as a technocrat who came into office believing the civil service already had the answers but was prevented from carrying them out by incompetent and malicious Conservatives. I do not really see the comparisons with Rishi and John Major. What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
Radical in this sense meaning "exactly the same as all Labour chancellors since at least the 80s, pretending to be fiscally responsible whilst borrowing and spending money like a drunken sailor".
‘Investment’ is one of those terms, like ‘defensive targets’ in military conflicts, can mean whatever people want it to.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
One of Mrs Foxy's cousins did a PhD in AI in the financial services industry at Southampton Uni, about 10 years ago, then went to work in Switzerland before moving back to London. He is very well paid and seems to be very valued by his bank.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
A PhD is totally different from a taught Masters. Southampton is quite a good research university so doesn't surprise me somebody getting a PhD from there ends up in a decent job. But its also worth noting, the world is very very different now. LLMs have completely disrupted academic research, even in the past 6 months, the world of doing research has completely shifted. Even in the past few weeks, the new ChatGPT model is now for the first time actually producing novel maths proofs* that academics are actually publishing and been peer reviewed as novel.
* there were claims upto a year ago of this, but were found that papers already existed but just unknown to those making the claims.
LLMs have not completely disrupted academic research. LLMs are proving interesting tools in one area of maths research and helpful tools at speeding up searching literature and writing code moe generally.
I was actually at dinner with some medical researchers in the US a couple of months ago. All they talked for over 2hrs about was how Claude Code had revolutionised their research and massively expanded their output.
If you don't think its disrupting much, and its just a bit of niche maths and some coding, I absolutely disagree. The number of submission to major conference in Computer Science also show evidence of this. There has just insane growth in the submissions, particularly the past year. They are having to cap the number of papers individuals can submit. I speak to computer scientists, mathematicans, physicist and all they tallk about this how Claude Code and Codex are integral part of their research now.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
btw within Posix you can choose what variety of Claude meets your need/budget. Opus is available.
Why would you do that, rather than just use Claude Code and use an MCP. All these companies trying to charge £20-30pm for each app, doesn't make a lot of sense when you can just install Claude Code and either via Cowork or MCP control the apps.
Because I don't have the nouse in that particular area, nor the time to work it up when there are other things to do
It is really easy. Claude Code will basically do all the setting up for you.
F'ing hell. Just installed Cowork and asked a few PB type questions. Demolishes my arguments within seconds and constructs new approaches.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Sorry but wanting the tax commitments funded is absolutely the right thing to do.
I applaud this move from Burnham. It shows some listening is going on and it helps a sector the SKS regime kicked repeatedly.
But wanting spending funded and knowing where it is coming from is the right thing to ask.
How does online distribution ‘cheat’ tax ? That’s a serious allegation.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Good morning, everyone.
Mr. 59, thanks for reminding us all how great and super awesome Prime Minister Burnham is, unlike those incompetent Tories.
Some might say we should wait more than three days before judging his performance and wait for actions over pronouncements but they're just flibbertigibbets.
Then in January 2026, the government announced pubs and music venues in England would be given a 15% discount on their business rates bills from April and would not see increases for two years.
Is Andy's another 20% ontop, or simply going from 15% to 20%?
January 2026 surely meant April 2026.
We are now in July so Burnham is talking about April 2027 so it's an extra discount.
Good
Hopefully businesses can hang on till then. Especially as the energy crisis is now escalating and those costs will be going up considerably in time for winter
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
If you are talking property taxes how is online distribution cheating business rates.
Vapes are straightforward because you just remove small business relief from some business types but you rapidly get into problems once you go to your other personal pet hates because all of them have legitimate competitors which you would also hit.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
Yes. Claude weights have leaked to China in the past, but this particular model it appears they did some mass distillation which you can't really do via simple just firing up the web interface and asking questions.
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
You can quibble about the details, but the fact remains that China is the only country other than the US with anything approaching a competitive sovereign AI program.
And while they are a few years behind in technology, they also have a formidable and growing chip industry.
Trump is helping them by making it much more difficult for Chinese who have trained in the US to stay in the US. It's perhaps unlikely that Yang Zhilin would have remained in the US under a different president, but there are many more Chinese researchers and entrepreneurs who would.
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Good morning, everyone.
Mr. 59, thanks for reminding us all how great and super awesome Prime Minister Burnham is, unlike those incompetent Tories.
Some might say we should wait more than three days before judging his performance and wait for actions over pronouncements but they're just flibbertigibbets.
Oh FFS…the right on here were judging his performance as “the worst” before he’d even started. Don’t rattle your pearls at someone on the left cheerleading him. Some people…
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Good morning, everyone.
Mr. 59, thanks for reminding us all how great and super awesome Prime Minister Burnham is, unlike those incompetent Tories.
Some might say we should wait more than three days before judging his performance and wait for actions over pronouncements but they're just flibbertigibbets.
Oh FFS…the right on here were judging his performance as “the worst” before he’d even started. Don’t rattle your pearls at someone on the left cheerleading him. Some people…
Mr. Seal, I've been broadly positive about Burnham. I was pleased to see his comments on the North Sea and axing the digital ID nonsense.
But getting carried away when, essentially, he's said a few things over a few days is a bit ridiculous. He's neither the Messiah nor a naughty boy. He hasn't had time to be either.
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
Yes. Claude weights have leaked to China in the past, but this particular model it appears they did some mass distillation which you can't really do via simple just firing up the web interface and asking questions.
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
You can quibble about the details, but the fact remains that China is the only country other than the US with anything approaching a competitive sovereign AI program.
And while they are a few years behind in technology, they also have a formidable and growing chip industry.
Trump is helping them by making it much more difficult for Chinese who have trained in the US to stay in the US. It's perhaps unlikely that Yang Zhilin would have remained in the US under a different president, but there are many more Chinese researchers and entrepreneurs who would.
China are following their now standard playbook. They decide they want to get good at something, they throw huge resources at it and are willing to steal IP and break whatever rules are stopping them. But, that only works because they also have extremely smart, dedicated and motivated people, and the government sets them up in vicious competition. And they go from copying to innovating. Deepseek have introduced several very important ideas that are not just nicked from somebody else.
R&D labs at likes ByteDance, Baidu, Tencent, Alibaba, Deepseek all doing loads of really good research along with some very good (and massive) university labs.
The EU total competitor is Mistral is just not in the game. UK does have top tier labs in DeepMind and some new labs that are getting funded from people who worked at DeepMind. To the credit of the government they are also putting some decent money aside for Gen AI.
If it's the sort of help that Britain got, it basically means KSA paying a great deal of money for nukes that depend on the USA and that they can't really use. That's safer for the USA than not helping them.
Wrong.
The renewed cooperation between the US and U.K. was a merging of two functioning nuclear programs.
The U.K. had multi-stage weapons that worked.
The result was a program where the designs were joint - US devices contain U.K. ideas and innovations. The U.K. was working on spherical secondaries and fission secondaries. This then led to the later designs (joint) for oblate secondaries.
The design teams merged, to an extent. The US and U.K. build their own flavours of the joint weapons designs.
The actual manufacturing is done in each country and is independent of each other.
You seem to be referring to Trident, which is the launch vehicle. And is entirely separate to the warheads.
Annoying discussion by Tom McTague and Oli Dugmore, discussing Starmer. You should imagine a Viewcode bouncing up and down on the chair shouting "No! You don't get it! You idiots".
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
From Starmer's early days as Prime Minister, I was forecasting that he would retire early like Wilson. He often made slips, and in recent speeches seems to be reading if not word-by-word, then certainly with frequent glances at his notes. Aside from that, I see Starmer as a lawyer, not a politician; as a technocrat who came into office believing the civil service already had the answers but was prevented from carrying them out by incompetent and malicious Conservatives. I do not really see the comparisons with Rishi and John Major. What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
Radical in this sense meaning "exactly the same as all Labour chancellors since at least the 80s, pretending to be fiscally responsible whilst borrowing and spending money like a drunken sailor".
Unlike the recent 14 years of Tory chancellors who have run such a tight ship that borrowing is at an all time low?
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
Sorry but wanting the tax commitments funded is absolutely the right thing to do.
I applaud this move from Burnham. It shows some listening is going on and it helps a sector the SKS regime kicked repeatedly.
But wanting spending funded and knowing where it is coming from is the right thing to ask.
How does online distribution ‘cheat’ tax ? That’s a serious allegation.
From the Government press release, so that we are fully informed: "Today’s changes will be fully funded, including through reviewing reliefs for businesses that do not make a positive contribution to local communities, such as vape shops. The Government will also crack down on businesses that sell through online marketplaces but do not comply with their tax obligations, putting them at an unfair advantage over businesses that play by the rules. The Government is consulting on measures to make online marketplaces more responsible for preventing non-compliant sellers from avoiding their tax obligations and further detail will be set out in due course. Revenue raised from these reforms will be reinvested in improvements to the business rates system."
That seems clear enough about where the funding is intended to come from, it's not reasonable to ask for precise details when the PM is only 3 days into the job, providing that the source of funding is reasonably plausible (which in this case it seems to be). The approach is to make the headline announcements now for political impact, say in principle how they will be paid for, and flesh out the detail in the coming weeks.
I doubt the 2034 election will be relevant to me to be fair but maybe I will have my pacemaker battery changed if required !!!
On the bus fares, Burnham in front of a bus with the £2 announcement shown prominently doesn't show it does not come in to 2027
Also seems there is the small print as ever:
Sky states further details of bus fare cap 'eligibility' still to be ironed out
Prime Minister Andy Burnham announced this morning he will cap single bus fare routes in England outside of London at £2 for all of 2027.
But it's not completely clear exactly which services this will apply to.
Lots of routes are excluded from the current £3 fare cap, according to gov.uk.
The Department for Transport said they would share more details in due course, but pointed out that under the £3 cap only open services that allow any member of the public to board, as well as services that allow free travel for the whole of the route, are eligible.
That would exclude dedicated school bus services.
On a visit to a bus depot in Bath to promote the policy, Burnham appeared in front of a branded bus with an asterisk that read "subject to local operator opt in".
*maximum fare, subject to local operator opt in
That sounds perfectly reasonable to me. I wouldn’t expect to jump on an Ember bus to Inverness for £2.
Fares for me are free though I never use a bus.
A man possessed of too much time a good tolerance for discomfort can use a Stagecoach day pass to visit Hereford from Oxford or vice versa for £5. Or £4 if purchased as a carnet of five. And these are no expresses - dozens and dozens of stops.
Stagecoach don’t run the Oxford-Cheltenham bus any more so I suspect that might not work.
I don't know the answer to this. Do "vape shops" need a licence from the council to be one? Because otherwise, how do we decide what a vape shop is? Any shop that sells vapes? What if I claim to be a cornershop supplying food, but just happen to have a rather large selection of vapes and in reality don't flog that many Mars bars?
30s take – looks like it is aimed at managers within large companies. If you want to use AI, just do it. A non-techie friend spends £10 a month on (probably) ChatGPT. Several PBers use AI including @ohnotnow in this very thread. Try self-hosting if you have the budget for a beefy-ish machine. Mac Minis were popular; a friend was looking to throw a few thousand HP's way:-
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
Totally Agree. I would just sign up for Claude Code or Codex (as in industry that is what everybody uses) and spend ages working with it to do all sorts of tasks. Learn about Skills and MCPs. Learn about what they do well and what they do poorly, when and how they break down, and how to guide them away from the poor responses. There is quite a few more than tricks to getting them to work effectively for you.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.
(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
Good luck with that. But on your statistical R work you must know about running Posix (=Claude Sonnet 4.6) in R. I just want to say that my experience of it has been v. good but it is pricey (£24 pm) and I'm not keen on it writing and running code directly on my Mac. Yesterday I tried the new Chinese AI Moonshot Kimi K3 on one of the same problems and it was at least as good and honed in on the solution quicker and sharper than Posix, for free. Guess whether my Posix sub will be renewed! The Chinese provenance might be an issue but the Californian techbros are in the crosshairs of a real competitor.
Kimi K3 is distilled Claude weights. Claude Sonnet is also significantly inferior to Opus let alone Fable. Fable is the bonkers, here is some random shitty data, can you do something with that, and off it goes and does some quite amazing things with little to no guidance.
"Kimi K3 is distilled Claude weights."Is that via industrial espionage ?
Yes. Claude weights have leaked to China in the past, but this particular model it appears they did some mass distillation which you can't really do via simple just firing up the web interface and asking questions.
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
You can quibble about the details, but the fact remains that China is the only country other than the US with anything approaching a competitive sovereign AI program.
And while they are a few years behind in technology, they also have a formidable and growing chip industry.
Trump is helping them by making it much more difficult for Chinese who have trained in the US to stay in the US. It's perhaps unlikely that Yang Zhilin would have remained in the US under a different president, but there are many more Chinese researchers and entrepreneurs who would.
China are following their now standard playbook. They decide they want to get good at something, they throw huge resources at it and are willing to steal IP and break whatever rules are stopping them. But, that only works because they also have extremely smart, dedicated and motivated people, and the government sets them up in vicious competition. And they go from copying to innovating. Deepseek have introduced several very important ideas that are not just nicked from somebody else.
R&D labs at likes ByteDance, Baidu, Tencent, Alibaba, Deepseek all doing loads of really good research along with some very good (and massive) university labs.
The EU total competitor is Mistral is just not in the game. UK does have top tier labs in DeepMind and some new labs that are getting funded from people who worked at DeepMind. To the credit of the government they are also putting some decent money aside for Gen AI.
The problem with AI is that so much of the original content was stolen (or if you want to be polite and ignore how it was sourced just used without permission) that I think most people agree that stealing from OpenAI / Anthropic is totally fair...
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
It's going to be a hell of a lot of increased taxes to pay for all these promises.
Brace ....
£100m...petty cash
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
If you are talking property taxes how is online distribution cheating business rates.
Vapes are straightforward because you just remove small business relief from some business types but you rapidly get into problems once you go to your other personal pet hates because all of them have legitimate competitors which you would also hit.
In addition, there are two games being played.
One is the use of business premises as a washing machine. The tenants, who conduct various illegal enterprises, pay a very high rent. Which is actually paying their boss. All the taxes are paid - because that buys legitimacy for the laundered money.
The second is popups in empty shops that then go out of business when taxes are due.
Some attempt to combine the two - but that is a bit obvious, even for Plod.
Comments
Burnham in first test of his resistance to rightwing media pressure with decision on ERS response to Conservative caused prisons crisis.
LOL!!
Sorry. It won't wash.
It's the only experience for getting the job.
Applying twice. Being rejected. Hasn't fucked the application up a third time.
Has actually executed a takeover almost no one thought possible.
Doesn't make him better at the job.
But it sure as hell made him better at getting it.
The BBC has also seen evidence the defence did request the maternity notes but they were not released.
1.01 x 10⁷ is in the tens of millions, just as the decade the 2010s was 2010 to 2019.
1 million is exactly 1,000,000
1,000,001 is in the millions.
https://youtu.be/q614DWF4OiM
Ancidata on this - I worked with quite a lot of Poles doing fairly low skilled factory work between 2008-12. Not all, but quite a lot of them are back in Poland now.
So yes, Yusuf is correct. Annoying, self-pitying, aggressive to others, whiny to himself, incapable of forming an argument without complaining, somebody who should not be let anywhere near power, yes, but in this he is correct (assuming the AI is right).
Also note this is gross inward movement, not net. After taking gross outward movement into account, the net inward movement will be smaller. But - to belabour the point - we're back to arguing about the side of the bus again.
China quietly saved the world last month
https://www.youtube.com/watch?v=BkA0bkb6ZO0
HP ZBook Fury G1i Mobile Workstation
Engineered for your most demanding workflows, offering desktop class performance: rendering, VFX, machine learning, AI inferencing, modelling, and more.
https://www.hp.com/gb-en/shop/products/laptops/hp-zbook-fury-g1i-18-inch-mobile-workstation-pc-wolf-pro-security-edition-data-science-8s9v1ea-abu
And as ever, there will be YouTube videos for guidance.
Health warning: I'm in the still thinking about finding time to think about it phase.
The author, the algorithm and the end of literary struggle
AI-generated texts flooding Amazon should scare anyone who cares about the written word" (£)
https://www.telegraph.co.uk/news/2026/07/21/author-algorithm-literary-struggle-ai-books-texts-amazon/
This is a pretty good and interesting article imo.
Sean Thomas
The author, the algorithm and the end of literary struggle
AI-generated texts flooding Amazon should scare anyone who cares about the written word
https://www.telegraph.co.uk/gift/9260b5afa2ab3a08
The pressure was intense, the stakes were high, and failure was not an option.
They are saying he's a basically decent man who messed up. I'm saying he was a lying bastard [rudeword] who bluffed his way into the top job and got spectacularly found out. They think he's John Major. I think he's criminally irresponsible and someone who was constitutionally incapable of performing the duties of the PM. I think he was mentally deficient, @Luckyguy1983 thinks he was evil, but either way he shouldn't have been let thru the door of number 10 and it speaks badly of the Labour Party that they let him stay in post so long.
Grrr...
Anyhoo, make your own mind up.
https://www.youtube.com/watch?v=PT756cTDG3k (49 mins)
https://www.bbc.co.uk/news/articles/c2lqd72d4lqo
They only came in as a temporary measure after the terrorist attacks on various airports. Andy B, come on, give the greedy bastard airport a kicking and force them to go back to nought or £1.
https://www.bbc.co.uk/news/articles/c2lqd72d4lqo
They only came in as a temporary measure after the terrorist attacks on various airports. Andy B, come on, give the greedy bastard airport a kicking and force them to go back to nought or £1. Would be really popular.
https://vf.politicalbetting.com/discussion/comment/5589664/#Comment_5589664
"...Get yourself a $5 VPS. Put a web based SSH server on it. Login. Install Claude Code or Codex or whatever. The cheapest way to get started is probably OpenCode + one of the really cheap Chinese models on the backend..."
Which is detailed and probably intended to be helpful, but it had me dribbling like somebody who's had a stroke. I don't know how to do any of that.
Just typing in here is some text, do som'it with it, won't get your very far.
I would be extremely wary of universities offering courses at the moment. I talk to a lot of academics and a lot of them are behind the curve of industry and their post-docs / PhDs and the landscape shifts fast. I don't think universities even really know how to run the traditional courses now that AI is so good.
There will be videos on YouTube offering guidance, although I cannot now find the one I was looking for.
I think FrancisUrquhart has posted here about using AI at work. ETA and now I see he has just posted.
- https://platform.claude.com/ enables you to "do" Claude Code without installing anything (hopefully)
- I don't think there is a remote account for Codex - you have to install it - but you can have an OpenAI account.
- So I'll start there
Once I've sorted that out, then I can do the download-install-run thing.(narrator: Viewcode is trying to analyse tens of thousands of documents (hundreds of thousands of documents). Since AI is the only thing that can realistically do that now that PERL is with nana and the angles, I have to work it out)
What was more interesting in the video was the notion that Rachel Reeves was a more radical Chancellor than she made out – changing fiscal rules to raise more money for investment.
I've just finished listening to Tom McTague and Oli Dugmore (https://www.youtube.com/watch?v=PT756cTDG3k ) and shouting at the screen, but one of the things that enabled me to get a handle on their discussion was when they mentioned "the Edgerton book". The one they meant was "The Rise and Fall of the British Nation: A Twentieth-Century History", which I enjoyed even though it's not his best book.
But the fact that they mentioned "the Edgerton book" meant that they and I had some common conceptual ground which in turn meant that I could theoretically join the discussion if invited. What do we - what do I - do when we don't have books in common?
"How, pray tell, does one "slap on Claude Code"?"
(Narrator. Viewcode is not posh)
https://code.claude.com/docs/en/quickstart
You run the relevant command, and it guides you through the setup. It is basically 2 steps. And then once its installed and your account verify, you can just type claude in any directory etc and then you will be in the world of Claude Code.
Its works far better with Linux / Mac than WIndows, because of the much more powerful command line terminals.
https://code.claude.com/docs/en/quickstart
You run the relevant command, and it guides you through the setup. It is basically 2 steps. And then once its installed and your account verify, you can just type claude in any directory etc and then you will be in the world of Claude Code.
https://www.koreatimes.co.kr/entertainment/shows-dramas/20260723/agent-kim-reactivated-airs-k-dramas-1st-fully-ai-scene
...People often assume AI content is created at the click of a button, but it took two full months to complete the rough cut of the three-minute video.
"We spent an entire month generating concept images for the motorcycle and car models, the tunnel and bridge, the surrounding mountains and even the ice floating on the river," Ryu said.
While the team delivered fewer than 300 final shots, they generated more than 2,000 during the production process. Ryu stressed that human expertise remains crucial.
"No matter how realistic the images look, viewers notice when the flow feels wrong," he said. "We approached it as if we were shooting live action, with particular attention to constructing and directing each shot using the visual grammar of cinema."
While the industry remains cautious, Ryu said an increasing number of creators are exploring AI video following the high-profile experiment with "Agent Kim Reactivated.".
That all been said, he ain't in the same universe of lying, cheating, only in it for themselves as Boris. Starmer was a level 1 to Boris level 100. And wasn't anywhere near as bad as Blair or Brown in terms of lying, dark arts, etc.
So Southampton have a bit of a track record in the field, not just jumping on the bandwagon.
* there were claims upto a year ago of this, but were found that papers already existed but just unknown to those making the claims.
https://www.washingtonpost.com/business/2026/07/22/us-nuclear-deal-with-saudis-includes-path-enriching-fuel/?utm_campaign=&utm_medium=email&utm_source=newsletter
https://www.youtube.com/watch?v=vgA0CE7XP2k
Andy Burnham just delivered his first speech as Prime Minister. There is one pattern hidden in it. Once you see it, you cannot unsee it.
Watch the pronouns.
Every promise that can visibly fail starts with 'I.' 'I will bring forward a plan.' 'I can give people breathing space.' 'I will set out measures tomorrow.' 'I will issue my first instruction.'
Then the moment the goal gets bigger than one person, he flips to 'we.' 'We will change politics.' 'We will take power out of here.' 'We will build more council homes.'
Then the sentence the whole speech has been building towards: 'I will give this my all, and I ask you all to pull with me.' I takes the responsibility. We does the work. That line is the leadership contract.
Most leaders get this backwards. They say 'we' when things go wrong and 'I' when things go right. Burnham flips it. That is how you ask people to trust you before you have earned it.
https://www.youtube.com/shorts/gt2Ofzye4WM
Burnham announces 20% cut in business rates for pubs, clubs and live music venues in England
Downing Street says the tax cut, which will cost around £100 million a year, will be funded by a review of tax reliefs currently given to businesses which "do not make a positive contribution to local communities, such as vape shops".
https://www.bbc.co.uk/news/live/cdewkelwrgjt
Betting shops too?
I think they didn't quite realise this. They were ending the big COVID discount and I don't think they appreciated that for lots of hospitality (already getting whacked by higher minimum wage and NI rises) they were plenty of case it lead to crazy rises in their business rates.
My Aunt used to run a pet shop in the shopping centre too in the late seventies and early eighties and got married in chelmsley wood.
It’s bizarre that he has this weird idea that travel writing will escape the oncoming Armageddon for writers. Travel is an area where there’s so much information available online that AIs would find creating travel articles easy peasy.
The Omani lane is dead now.
Next escalation step will be for Iran to block Fujairah.
More escalation seems likely.
https://x.com/hfi_research/status/2080157780397408766?s=61
Of the Chinese labs, DeepSeek seem to be the ones who are making genuine algorithmic advances, doing some very clever innovations. Their backstory is still bullshit, it isn't a handful of quant guys who decided they would give LLMs a bit of a spin. It a huge team of very smart people with access to the banned Nvidia cards. But some of these other Chinese labs are much more dodgy.
Interestingly, also 28 tanks and 12 armoured fighting vehicles - biggest totals in a long while. Looks like the the generals were told to make a breakthrough. Ukraine was waiting.
The lack of attached conditions (in contrast to the deal being negotiated by the last administration) doesn't help.
Is Andy's another 20% ontop, or simply going from 15% to 20%?
Brace ....
We are now in July so Burnham is talking about April 2027 so it's an extra discount.
In October 2024, the account on Polymarket registered in the name “GCottrell93” received about $8.8mn from two separate wallets. The funds were deposited from two crypto wallets and then used to bet on Donald Trump winning the US presidential election.
Shortly after the election result, which generated $4.46mn in profit, almost all of the money in the account was emptied into two different wallets. It is unclear who provided the money used by the account and who ultimately received the winnings.
https://bsky.app/profile/snellarthur.bsky.social/post/3mrc3djdf4k27
Will be funded by increased vigilence and collection from LESS worthy recipients like vape shops, turkish barbers, possibly betting shops you'd imagine and out of town warehouses and online distribution cheating taxes.
small steps make hard miles.
The Right Wing panic is frankly hilarious. Wanting tax commitments priced to the penny after Boris, Truss and Sunak and the Kemiquake bulshit that everythis will be paid for by utterly abolishing net zero are returning the UK to a 20th century energy State Coal / Gas / OIl add in a bit of Fracking!
If you don't think its disrupting much, and its just a bit of niche maths and some coding, I absolutely disagree. The number of submission to major conference in Computer Science also show evidence of this. There has just insane growth in the submissions, particularly the past year. They are having to cap the number of papers individuals can submit. I speak to computer scientists, mathematicans, physicist and all they tallk about this how Claude Code and Codex are integral part of their research now.
I applaud this move from Burnham. It shows some listening is going on and it helps a sector the SKS regime kicked repeatedly.
But wanting spending funded and knowing where it is coming from is the right thing to ask.
How does online distribution ‘cheat’ tax ? That’s a serious allegation.
Mr. 59, thanks for reminding us all how great and super awesome Prime Minister Burnham is, unlike those incompetent Tories.
Some might say we should wait more than three days before judging his performance and wait for actions over pronouncements but they're just flibbertigibbets.
Hopefully businesses can hang on till then. Especially as the energy crisis is now escalating and those costs will be going up considerably in time for winter
Vapes are straightforward because you just remove small business relief from some business types but you rapidly get into problems once you go to your other personal pet hates because all of them have legitimate competitors which you would also hit.
And while they are a few years behind in technology, they also have a formidable and growing chip industry.
Trump is helping them by making it much more difficult for Chinese who have trained in the US to stay in the US.
It's perhaps unlikely that Yang Zhilin would have remained in the US under a different president, but there are many more Chinese researchers and entrepreneurs who would.
But getting carried away when, essentially, he's said a few things over a few days is a bit ridiculous. He's neither the Messiah nor a naughty boy. He hasn't had time to be either.
R&D labs at likes ByteDance, Baidu, Tencent, Alibaba, Deepseek all doing loads of really good research along with some very good (and massive) university labs.
The EU total competitor is Mistral is just not in the game. UK does have top tier labs in DeepMind and some new labs that are getting funded from people who worked at DeepMind. To the credit of the government they are also putting some decent money aside for Gen AI.
The renewed cooperation between the US and U.K. was a merging of two functioning nuclear programs.
The U.K. had multi-stage weapons that worked.
The result was a program where the designs were joint - US devices contain U.K. ideas and innovations. The U.K. was working on spherical secondaries and fission secondaries. This then led to the later designs (joint) for oblate secondaries.
The design teams merged, to an extent. The US and U.K. build their own flavours of the joint weapons designs.
The actual manufacturing is done in each country and is independent of each other.
You seem to be referring to Trident, which is the launch vehicle. And is entirely separate to the warheads.
Er…
"Today’s changes will be fully funded, including through reviewing reliefs for businesses that do not make a positive contribution to local communities, such as vape shops.
The Government will also crack down on businesses that sell through online marketplaces but do not comply with their tax obligations, putting them at an unfair advantage over businesses that play by the rules.
The Government is consulting on measures to make online marketplaces more responsible for preventing non-compliant sellers from avoiding their tax obligations and further detail will be set out in due course. Revenue raised from these reforms will be reinvested in improvements to the business rates system."
That seems clear enough about where the funding is intended to come from, it's not reasonable to ask for precise details when the PM is only 3 days into the job, providing that the source of funding is reasonably plausible (which in this case it seems to be). The approach is to make the headline announcements now for political impact, say in principle how they will be paid for, and flesh out the detail in the coming weeks.
I should get out more.
One is the use of business premises as a washing machine. The tenants, who conduct various illegal enterprises, pay a very high rent. Which is actually paying their boss. All the taxes are paid - because that buys legitimacy for the laundered money.
The second is popups in empty shops that then go out of business when taxes are due.
Some attempt to combine the two - but that is a bit obvious, even for Plod.