Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
But Incorrect Horse Battery says Claude Code is rubbish. I think it is probably more you aren't using them very well. Are they 100% correct, no, but humans makes masses of mistakes. Is Claude Code + Opus absolute elite tier coder, yes.
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
The people who were purely “code monkeys” — i.e. implementing tickets with minimal understanding of the system — were already in a fragile position. AI just makes that fragility obvious. But that role was brittle long before LLMs showed up. It was vulnerable to offshoring, automation, better frameworks, you name it.
Indeed. The truth is a lot of people currently making a living writing code are not particularly skilled or valuable, the work they do is not hard. Some of them will either be replaced by LLMs or demoted to just bug-fixing the output.
I write mostly Verilog and assembly language and design much of the hardware that hosts my code. I do not expect to be replaced by an LLM any time in the foreseeable future. But if I spent my days writing Python or maintaining some shonky mobile app, I'd be worried.
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
But Incorrect Horse Battery says Claude Code is rubbish. I think it is probably more you aren't using them very well. Are they 100% correct, no, but humans makes masses of mistakes. Is Claude Code + Opus absolute elite tier coder, yes.
AI makes masses of mistakes and this latest iteration is no better.
You have your experiences, I have mine. But allowing Claude into our organisation has been a really, really bad move and I can only speak to that.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
This is my exact experience. I used all the LLMs via Chat Window or via Cursor etc and they were ok but I used them for boilerplate and basically fancy autocomplete and sped up things a bit. But Claude Code + Opus model came out, and it was a huge step change. I have written more code in the past 6 weeks than I would be able to manage in 6 months and this is implementing state of the art ML algorithms many of which don't have public code repos.
The secret is planning. You spend time chatting through the plan. As long as you do that, and you go step by step through what the plan is, you don't just do "code me this", it is super human level coder. Its like have a team of really really good devs.
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
But Incorrect Horse Battery says Claude Code is rubbish.
It is.
It created a bug I’ve been fixing literally today
One of our juniors is obsessed with it.
Right.
So you merged AI-generated code into something important without fully understanding it, without adequate tests, and without catching the bug in review… and the conclusion is “the tool is rubbish”.
That’s like blaming Visual Studio because you copy-pasted from Stack Overflow and prod fell over.
The obvious error isn’t in Claude. It’s in the workflow.
If a junior is “obsessed with it” and generating bugs, that’s not an AI problem — that’s a supervision and engineering standards problem. You don’t let juniors ship unchecked code whether it came from their own brain or a stochastic parrot.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Aditya Agarwal is absolute elite level coder, early algos for Facebook, then Dropbox,
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
But Incorrect Horse Battery says Claude Code is rubbish.
It is.
It created a bug I’ve been fixing literally today
One of our juniors is obsessed with it.
Right.
So you merged AI-generated code into something important without fully understanding it, without adequate tests, and without catching the bug in review… and the conclusion is “the tool is rubbish”.
That’s like blaming Visual Studio because you copy-pasted from Stack Overflow and prod fell over.
The obvious error isn’t in Claude. It’s in the workflow.
If a junior is “obsessed with it” and generating bugs, that’s not an AI problem — that’s a supervision and engineering standards problem. You don’t let juniors ship unchecked code whether it came from their own brain or a stochastic parrot.
He didn’t check in the code. I found it during a code review, I’m fixing it on his behalf because after 8 iterations of pushing to GitHub where he’s obviously just got it into a loop I’ve taken over. Luckily I ran the code locally, otherwise it would have gone in and I’d be working late tonight.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
Microsoft are our second largest client and so one of the C-suite here was desperate for CoPilot to be used.
I used it out of interest for some basic analytics of a spreadsheet I’d made and it added two numbers incorrectly.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini in general pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Its the enabling of the context in terminal / IDE plus some secret sauce that Cluade Opus currently has that makes it stand out. I hear things Kimi are catching up though.
I was irritated this morning; Financial Advisor's team wanted to 'discuss progress' with us and use Teams. Unfortunately (!) my Mac Catalina, which works perfectly well with everything else I want to use, including Zoom, isn't acceptable to Microsoft's Teams. I tried with newer iPad, but that required installation of the Teams App, and by the time I realised the situation, time was running out.
To be fair, I was already irritated because at 3am I was woken by by a text from my bank telling me that they'd been suspicious of, and therefore blocked, a payment and sorting it out when I was finally properly conscious took quite a long time. Particularly because said payment was a scam of some sort. Anyone heard of AQ*MAIDS? I haven't. So that's my credit card out of action for three or so days!
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Gemini pretty good hahahahaha.
Like when it told me multiple times that Angular does not support @if syntax.
Correction: A previous version of this post misidentified the South Caucasus as belonging to Russia. The region, made up of territory in Armenia, Georgia and Azerbaijan, is not part of Russia. We deleted the previous tweet.
Is HY writing for the Post now?
In fairness to @HYUFD , he has never(?) stood on the side of the Russians, and WaPo has been reduced to the status of used toilet paper by i) Trump's sucking up to Putin and ii) Bezos's sucking up to Trump, not through the influence of a English councillor-candidate and Conservative party member, no matter how many tanks the latter may have.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
Ryan Air, apparently...
Nobody doing anything serious with AI uses copilot, I'm using Cursor with a very carefully crafted set of rules, domain knowledge and memories.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
Microsoft sacked the majority of their QA team almost a decade ago. The users do the QA testing now.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Gemini pretty good hahahahaha.
Like when it told me multiple times that Angular does not support @if syntax.
A bit like girl who says whenever I date I always seem to end up with arseholes, I feel like you might be the one using these tools wrong.
Some of the most elite coders in the world are saying these tools, especially Claude Code is now top tier. But you seem to have a terrible experience e.g. are you just using free plans via web interface or are you using the max tier models with CLI integration?
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
People actually believe this shit
No wonder Lisa has referred Telegraph Mail merger.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
Ryan Air, apparently...
Nobody doing anything serious with AI uses copilot, I'm using Cursor with a very carefully crafted set of rules, domain knowledge and memories.
I’m assuming it has a massive user base by virtue of being forced on virtually every PC. But it is awful.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
There is a similar story with D365 where Microsoft last week introduced a bug into the Plan table that disabled all customisations.
Slight problem it's been in production for years and a lot of businesses use it for Project management and Field Service organisation..
It's definitely the case that Microsoft's quality control (which was never that great) is getting far far worse.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Gemini pretty good hahahahaha.
Like when it told me multiple times that Angular does not support @if syntax.
A bit like girl who says whenever I date I always seem to end up with arseholes, I feel like you might be the one using these tools wrong.
No it’s just overhyped and not very good. We aren’t going to agree on this but I am very confident in my views here.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Gemini pretty good hahahahaha.
Like when it told me multiple times that Angular does not support @if syntax.
A bit like girl who says whenever I date I always seem to end up with arseholes, I feel like you might be the one using these tools wrong.
Some of the most elite coders in the world are saying these tools, especially Claude Code is now top tier. But you seem to have a terrible experience e.g. are you just using free plans via web interface or are you using the max tier models with CLI integration?
Using it via the CLI and fully integrated into our code base using custom instructions and context. It isn’t good, end of story.
Our company pays for it because they’re convinced it’s the future (another reason I think it’s overhyped).
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
There is a similar story with D365 where Microsoft last week introduced a bug into the Plan table that disabled all customisations.
Slight problem it's been in production for years and a lot of businesses use it for Project management and Field Service organisation..
It's definitely the case that Microsoft's quality control (which was never that great) is getting far far worse.
On my work laptop if I click the shut down button to do an update 9/10 times it will not shut down but restart. Very odd. But this is like the third PC I’ve seen do this.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
People actually believe this shit
No wonder Lisa has referred Telegraph Mail merger.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Did anyone ever?
As your representative from the public sector, I can confirm that that is the AI tool which we have to work with, and it is at best mixed. I wouldn't do anything technical with it, and it has a tendency to give you the answer it thinks you want, rather than the correct answer.
There are huge variations now. Copilot shit. ChatGPT for code hmm, Claude via Chat Window spews way too much code and overcomplicates, Google Gemini pretty good, Claude Code in terminal / IDE with use of plan mode, elite.
Gemini pretty good hahahahaha.
Like when it told me multiple times that Angular does not support @if syntax.
A bit like girl who says whenever I date I always seem to end up with arseholes, I feel like you might be the one using these tools wrong.
Some of the most elite coders in the world are saying these tools, especially Claude Code is now top tier. But you seem to have a terrible experience e.g. are you just using free plans via web interface or are you using the max tier models with CLI integration?
Using it via the CLI and fully integrated into our code base using custom instructions and context. It isn’t good, end of story.
Our company pays for it because they’re convinced it’s the future (another reason I think it’s overhyped).
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar. I have already listed the top tier talent (that don't work for these companies so aren't shilling) who have publicly stated I am an elite level coder, that was my special talent, it isn't a special talent anymore.
Thinking of the governments response to some blokes views on immigration there must be some sort of itv3 show in it - Sir Keir Starmer and Dermot O'Leary tell you how you should feel about Dappy off of N Dubz views on Breakfast Clubs.
Former Norwegian Prime Minister Thorbjørn Jagland has been charged with "gross corruption" over his ties with the US sex offender Jeffrey Epstein, police have said.
Former Norwegian Prime Minister Thorbjørn Jagland has been charged with "gross corruption" over his ties with the US sex offender Jeffrey Epstein, police have said.
Former Norwegian Prime Minister Thorbjørn Jagland has been charged with "gross corruption" over his ties with the US sex offender Jeffrey Epstein, police have said.
An interesting comment on AI from a colleague of mine today. We are very much encouraged to use it at work, and there's no coding:
- Very senior people think it's amazing. Because they don't do much actual work, they delegate. This gives them plausible answers much quicker than asking someone, and certainly more so than if they tried to do the work themselves.
- Very junior people see AI as a shortcut where their knowledge or skills are lacking. It probably improves their output speed and quality. My fear is it also slows their learning on how to be better.
- Experienced, non-management staff find AI offers much more marginal gains. Because they are already very efficient at completing tasks, can send emails or do calculations or make presentations quickly and to a high quality without needing AI. AI helps, but its assistance is more marginal than people in the senior or junior group think it is.
The challenge is going to be how to train the experienced workers of the future if juniors are always taking short cuts via AI that does not yet exceed the abilities of experienced people... And how to stop management obsessing over it when its usefulness is limited.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
Yes, it's really just a very capable plagiarism machine.
An interesting comment on AI from a colleague of mine today. We are very much encouraged to use it at work, and there's no coding:
- Very senior people think it's amazing. Because they don't do much actual work, they delegate. This gives them plausible answers much quicker than asking someone, and certainly more so than if they tried to do the work themselves.
- Very junior people see AI as a shortcut where their knowledge or skills are lacking. It probably improves their output speed and quality. My fear is it also slows their learning on how to be better.
- Experienced, non-management staff find AI offers much more marginal gains. Because they are already very efficient at completing tasks, can send emails or do calculations or make presentations quickly and to a high quality without needing AI. AI helps, but its assistance is more marginal than people in the senior or junior group think it is.
The challenge is going to be how to train the experienced workers of the future if juniors are always taking short cuts via AI that does not yet exceed the abilities of experienced people... And how to stop management obsessing over it when its usefulness is limited.
I think this a decent take and it is human nature to take shortcuts / not check things. Also if you don't have any experience, you don't know what you don't know and LLMs provide information with extreme over confidence.
One issue I can foresee is companies cutting back on juniors (or expect them to churn lots more out), especially in things like coding, then in 10 years going shit we don't have any experienced staff because the ones we had are retiring and we didn't hire enough / they weren't trained up.
The only point I would disagree with is not that helpful for experienced. I am experience and it has made me massively more productive, but I can read / understand the technical stuff, make the plans with the LLM and check that it is going to do roughly the right thing, then can read the code quickly and understand what it does to weed out the mistakes. Also, I know what to ask it to check for to eliminate common mistakes. Along with the planning, the asking for LLM output to check themselves in the correct way is another crucial step.
An interesting comment on AI from a colleague of mine today. We are very much encouraged to use it at work, and there's no coding:
- Very senior people think it's amazing. Because they don't do much actual work, they delegate. This gives them plausible answers much quicker than asking someone, and certainly more so than if they tried to do the work themselves.
- Very junior people see AI as a shortcut where their knowledge or skills are lacking. It probably improves their output speed and quality. My fear is it also slows their learning on how to be better.
- Experienced, non-management staff find AI offers much more marginal gains. Because they are already very efficient at completing tasks, can send emails or do calculations or make presentations quickly and to a high quality without needing AI. AI helps, but its assistance is more marginal than people in the senior or junior group think it is.
The challenge is going to be how to train the experienced workers of the future if juniors are always taking short cuts via AI that does not yet exceed the abilities of experienced people... And how to stop management obsessing over it when its usefulness is limited.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
Microsoft sacked the majority of their QA team almost a decade ago. The users do the QA testing now.
Which starts to become a serious problem when you’re paying them tens of thousands of dollars per year in enterprise licences, and they fail at the very basic level.
Today I asked a question about contingency in a capital cost estimate and the AI tool conflated contingency and uncertainty, confidently telling me that the contingency was between -30% and +50%.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
I disagree, it's about having good guardrails, and above all context. Where you do have a point is the language you choose does matter in as much as the LLMs are trained on existing repos, the more common the language the "better" the training.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
I am involved with SOTA ML reasearch...so its the latter. And I don't find that, anymore. I am implementing things that either don't exist or only came out in past few weeks. But its obviously got loads of background "knowledge" of python / c++.
One big thing that Claude Code gives you and I have found the really mindblowing, I have seen it numerous times say to itself, I don't know or this is a new version that has new functionality, just go and download all the docs / tutorials / blogs, read them, then can use that new functionality.
Now if you want to tell me some code in Fortran or Ada to runs some nuclear power plant and 5 people ever seen that codebase, probably, it does need the background data knowledge.
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
nobody uses copilot anymore
Ryan Air do
Ryaniar’s copiliot is paying to be sitting there.
just lolled at that
Oh it’s very much true.
A Ryanair copilot is earning around £50 per flight hour, on a zero-hours contract, and has a £30k debt for their type rating that they need to pay off.
There’s caravan parks and car parks near Stanstead filled with FR first officers living in trailers and vans.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
Yes, it's really just a very capable plagiarism machine.
Or incapable. I used AI to produce a summary of a paper, and the summary referred to a person not mentioned in the paper.
I still don’t understand why we are all talking like life is just going to continue as it is
It’s utterly farcical. It’s like talking about the upcoming cricket season in February 2020
I assume this is related to AI..?
Or aliens.
AI Aliens.
I've been using LLMs for a while and agentic dev patterns in earnest since late last year. We're not about to be enslaved by our AI overlords. Change is coming and my fear is not the AI, but rather our government's complete inability to respond effectively
I think our erstwhile reporter from the front of Reddit believes two things. (1) AGI has been achieved and LLM's are conscious and (2) we are about to see hundreds of thousands or millions of white collar jobs vanish in a very short time.
In think (1) is wrong but there is something in (2) for sure. Maybe we are finally approaching the future of Tomorrow's World where no one has to work, and its just leisure all the time...
1 is wrong, absolutely nothing in it at all. That's not to say that LLM are improving seemingly exponentially, but they are nothing like AGI and not showing anything like consciousness. 2 is massively overblown
He is a believer in (1). I think we are seeing jobs taken by AI - call centres, translators etc. Time will tell
Jobs have been replaced by technology for centuries now. New jobs have been created by technology for centuries now.
They will continue to be.
Some will have to adapt. I would not want to be getting paid to write nonsense when AI can just generate it for instance.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
Yes, it's really just a very capable plagiarism machine.
And for a lot of applications, that's more than enough.
Good writers borrow, great writers steal, as I've just stolen.
We aren't going to agree, but I think you are a massive outlier e.g. I go to all the top academic conferences in things like ML, smartest minds on the planet, they basically all have Claude Code running 24/7. All the FAANG company employees I know are the same, if not Claude Code, something very similar.
I believe the differing experiences of AI coding often come down to what your use case is. If you're writing code to do something that's not unusual, using a popular language that has masses of publicly available source code to copy from, then the output can be useful. Step off that path and it's usefulness deteriorates sharply.
Yes, it's really just a very capable plagiarism machine.
And for a lot of applications, that's more than enough.
Good writers borrow, great writers steal, as I've just stolen.
This is where the West is particularly at risk, we have lots of jobs where we have educated people to a slightly higher level enabling them to do tasks that require some extra base knowledge, but in the limit are still quite repetitive and don't require huge amount of deep thought e.g. lots of coding, while the project manager role doesn't go away, it just expands. LLMs are already very good at this.
I am sure down the line new jobs come along but that isn't comforting for those in the firing line in 5 years time. Also, its was the West's advantage, this more educated workforce to do service jobs. That advantage is being swept away e.g. translation, you don't need it for general day to day things now. For Medical / Legal scenarios yes, but not for just contacting other people inquiring about things.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
Not necessarily. Actually *reading* the article it says...
It can now be revealed that the children taking part in the trial will be incentivised with the promise of up to £500 in vouchers for completing psychometric tests. These will measure the effect of the puberty blockers on their brains, including effects on impulse control and memory...Participants, aged under 16, will receive £30 vouchers for each of the 15 cognitive assessments they complete, as well as £15 vouchers for each of the three MRI scans they undertake, over the two years of the trial...Children in another arm of the trial, not taking the drugs, will be given £15 vouchers for each bone density scan and blood test they have.
Points to note
The children are already in the trial and have been randomised into one of the two arms, so they've already been consented.
Those in the intervention arm (on the drugs) are paid a nominal amount (£15 or £30) for each of the tests (cogntive assessments or MRI scans) they undertake
Those in the control arm (not on the drugs) are paid a nominal amount (£15) for each of the tests (bone density scan and blood test) they undertake
In short the participants are paid nominal sums to take part in tests they have already agreed to prior to assignment, and that this is an incentive to complete the trial, not an incentive to take part in the trial. Since a criticism of many studies in the trans sphere is dropout rate, this is a good idea.
There's a concept in RCT called "equipoise". The surgeons/doctors in a study must be genuinely undecided/unbiased as to which of the two arms is better, and the patients must be equally noncommittal: this enables randomisation to take place believably. In trans studies this is difficult since if the patients believe the study is biased or bent towards one outcome they will refuse to take part or simply bugger off to Europe and get the treatment they want there (Baroness Cass pointed this out as a reason to undertake the PATHWAY studies). The PATHWAY designers seem to have created a study that enables equipoise. The fact that these sums are paid to make sure they stay in the study prevents dropout
(Incidentally @isam, you said last night you wanted to know how tests of drugs in humans take place. You might want to google the term "CTIMP")
Angela Rayner is currently making some huge attacks on Rachel Reeves over business rates, energy bills and hospitality VAT.
Speaking at the nighttime economy summit in Liverpool, Rayner also appears to be doing some major pro-business positioning ahead of the inevitable leadership contest.
She says: "Confidence in politics matters. Businesses need to believe they will be treated fairly. That the rules won't shift without warning. That the long-standing structural issues will finally be addressed, not deferred again."
She says the government needs to drop ideology and be pragmatic to help businesses 👀
She is right though and sensible take on business
Meh. It's words. Does anyone seriously think Rayner's political instincts, or the interest groups she intends to serve, are pro-business ?
She pioneered the so-called workers rights bill at the behest of the Unions. A few token watering down of a couple of its elements doesn’t change that.
People might be missing the point. The most important part of what Rayner said is about stability: That the rules won't shift without warning. By and large, businesses (and people) can cope with or manage around policies they don't like, but they can't handle repeated, arbitrary lurches.
Labour politics-wise, this might be a repeat of the prawn cocktail offensive but more likely she has been talking to Andy Burnham.
She's more beholden to left wing ideology and has more populist instincts than Reeves (or Starmer, to the extent one can detect any coherent political philosophy or instincts in him at all). That is a recipe for more tax on business and more arbitrary changes, not less.
I like her, I'd want her fighting my corner if I was in her client group, but I'm not.
The focus on energy prices is an interesting one. Those are pretty much back at their 2021 prices in real terms this year, and the Govt will meet their manifesto reduction pledge.
If Starmer gets his f*cking comms sorted out (a very big if), it is potentially a winning issue.
(Forecast is 2026, not 2025. I looked over the numbers.)
Petrol is cheaper in cash terms, never mind real terms, than in 2012.
AEP of Telegraph reckons further to go as there is a global glut.
In my part of the world petrol is now cheaper than at any time offer the pandemic. 49p/litre for Super 98.
AEP might actually be right for a change.
I think he's almost certainly right: it's a combination of investment in oil and gas post Ukraine invasion, and continued roll out of renewables.
Things I would not want to be spending money on now: new nuclear plants.
If there’s a glut of oil that’s not great for some petrol states who need it to stay at a relatively high level.
That is absolutely correct.
It's a particular disaster for Russia, where production is being negatively impacted by lack of access to Western production equipment.
I still don’t understand why we are all talking like life is just going to continue as it is
It’s utterly farcical. It’s like talking about the upcoming cricket season in February 2020
I assume this is related to AI..?
Or aliens.
AI Aliens.
I've been using LLMs for a while and agentic dev patterns in earnest since late last year. We're not about to be enslaved by our AI overlords. Change is coming and my fear is not the AI, but rather our government's complete inability to respond effectively
I think our erstwhile reporter from the front of Reddit believes two things. (1) AGI has been achieved and LLM's are conscious and (2) we are about to see hundreds of thousands or millions of white collar jobs vanish in a very short time.
In think (1) is wrong but there is something in (2) for sure. Maybe we are finally approaching the future of Tomorrow's World where no one has to work, and its just leisure all the time...
1 is wrong, absolutely nothing in it at all. That's not to say that LLM are improving seemingly exponentially, but they are nothing like AGI and not showing anything like consciousness. 2 is massively overblown
He is a believer in (1). I think we are seeing jobs taken by AI - call centres, translators etc. Time will tell
Jobs have been replaced by technology for centuries now. New jobs have been created by technology for centuries now.
They will continue to be.
Some will have to adapt. I would not want to be getting paid to write nonsense when AI can just generate it for instance.
@leon should only really worry if the editor role is taken over by AI.
Jibes aside (apologies Leon) it is really quite challenging to work out quite which jobs are likely to succumb to AI. If you'd asked me some years ago I wouldn't particularly have said developers for example, and yet in hindsight its obvious in that all they do is write pretty much the same thing as other developers but with variations.
I don't know, but I do wonder, who cleans Waymo taxis?
AGI is a long way off, and none of the current 'AIs' can (in my view) ever result in such a thing. They're all on a very interesting path, but regurgitation isn't the way.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
Not necessarily. Actually *reading* the article it says...
It can now be revealed that the children taking part in the trial will be incentivised with the promise of up to £500 in vouchers for completing psychometric tests. These will measure the effect of the puberty blockers on their brains, including effects on impulse control and memory...Participants, aged under 16, will receive £30 vouchers for each of the 15 cognitive assessments they complete, as well as £15 vouchers for each of the three MRI scans they undertake, over the two years of the trial...Children in another arm of the trial, not taking the drugs, will be given £15 vouchers for each bone density scan and blood test they have.
Points to note
The children are already in the trial and have been randomised into one of the two arms, so they've already been consented.
Those in the intervention arm (on the drugs) are paid a nominal amount (£15 or £30) for each of the tests (cogntive assessments or MRI scans) they undertake
Those in the control arm (not on the drugs) are paid a nominal amount (£15) for each of the tests (bone density scan and blood test) they undertake
In short the participants are paid nominal sums to take part in tests they have already agreed to prior to assignment, and that this is an incentive to complete the trial, not an incentive to take part in the trial. Since a criticism of many studies in the trans sphere is dropout rate, this is a good idea.
There's a concept in RCT called "equipoise". The surgeons/doctors in a study must be genuinely undecided/unbiased as to which of the two arms is better, and the patients must be equally noncommittal: this enables randomisation to take place believably. In trans studies this is difficult since if the patients believe the study is biased or bent towards one outcome they will refuse to take part or simply bugger off to Europe and get the treatment they want there (Baroness Cass pointed this out as a reason to undertake the PATHWAY studies). The PATHWAY designers seem to have created a study that enables equipoise. The fact that these sums are paid to make sure they stay in the study prevents dropout
(Incidentally @isam, you said last night you wanted to know how tests of drugs in humans take place. You might want to google the term "CTIMP")
Why are people dropping out? If there is a good reason why they are, the trial should be reporting that as part of the data.
They should not be remaining in a trial they would otherwise drop out of, solely due to fiscal considerations.
I still don’t understand why we are all talking like life is just going to continue as it is
It’s utterly farcical. It’s like talking about the upcoming cricket season in February 2020
I assume this is related to AI..?
Or aliens.
AI Aliens.
I've been using LLMs for a while and agentic dev patterns in earnest since late last year. We're not about to be enslaved by our AI overlords. Change is coming and my fear is not the AI, but rather our government's complete inability to respond effectively
I think our erstwhile reporter from the front of Reddit believes two things. (1) AGI has been achieved and LLM's are conscious and (2) we are about to see hundreds of thousands or millions of white collar jobs vanish in a very short time.
In think (1) is wrong but there is something in (2) for sure. Maybe we are finally approaching the future of Tomorrow's World where no one has to work, and its just leisure all the time...
1 is wrong, absolutely nothing in it at all. That's not to say that LLM are improving seemingly exponentially, but they are nothing like AGI and not showing anything like consciousness. 2 is massively overblown
He is a believer in (1). I think we are seeing jobs taken by AI - call centres, translators etc. Time will tell
Jobs have been replaced by technology for centuries now. New jobs have been created by technology for centuries now.
They will continue to be.
Some will have to adapt. I would not want to be getting paid to write nonsense when AI can just generate it for instance.
@leon should only really worry if the editor role is taken over by AI.
Jibes aside (apologies Leon) it is really quite challenging to work out quite which jobs are likely to succumb to AI. If you'd asked me some years ago I wouldn't particularly have said developers for example, and yet in hindsight its obvious in that all they do is write pretty much the same thing as other developers but with variations.
I don't know, but I do wonder, who cleans Waymo taxis?
AGI is a long way off, and none of the current 'AIs' can (in my view) ever result in such a thing. They're all on a very interesting path, but regurgitation isn't the way.
One job where AI could probably surpass the human currently doing it is Prime Minister of the UK. Labour should consider it as an option for interim leader.
Angela Rayner is currently making some huge attacks on Rachel Reeves over business rates, energy bills and hospitality VAT.
Speaking at the nighttime economy summit in Liverpool, Rayner also appears to be doing some major pro-business positioning ahead of the inevitable leadership contest.
She says: "Confidence in politics matters. Businesses need to believe they will be treated fairly. That the rules won't shift without warning. That the long-standing structural issues will finally be addressed, not deferred again."
She says the government needs to drop ideology and be pragmatic to help businesses 👀
She is right though and sensible take on business
Meh. It's words. Does anyone seriously think Rayner's political instincts, or the interest groups she intends to serve, are pro-business ?
She pioneered the so-called workers rights bill at the behest of the Unions. A few token watering down of a couple of its elements doesn’t change that.
People might be missing the point. The most important part of what Rayner said is about stability: That the rules won't shift without warning. By and large, businesses (and people) can cope with or manage around policies they don't like, but they can't handle repeated, arbitrary lurches.
Labour politics-wise, this might be a repeat of the prawn cocktail offensive but more likely she has been talking to Andy Burnham.
She's more beholden to left wing ideology and has more populist instincts than Reeves (or Starmer, to the extent one can detect any coherent political philosophy or instincts in him at all). That is a recipe for more tax on business and more arbitrary changes, not less.
I like her, I'd want her fighting my corner if I was in her client group, but I'm not.
The focus on energy prices is an interesting one. Those are pretty much back at their 2021 prices in real terms this year, and the Govt will meet their manifesto reduction pledge.
If Starmer gets his f*cking comms sorted out (a very big if), it is potentially a winning issue.
(Forecast is 2026, not 2025. I looked over the numbers.)
Petrol is cheaper in cash terms, never mind real terms, than in 2012.
AEP of Telegraph reckons further to go as there is a global glut.
In my part of the world petrol is now cheaper than at any time offer the pandemic. 49p/litre for Super 98.
AEP might actually be right for a change.
I think he's almost certainly right: it's a combination of investment in oil and gas post Ukraine invasion, and continued roll out of renewables.
Things I would not want to be spending money on now: new nuclear plants.
If there’s a glut of oil that’s not great for some petrol states who need it to stay at a relatively high level.
That is absolutely correct.
It's a particular disaster for Russia, where production is being negatively impacted by lack of access to Western production equipment.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
Not necessarily. Actually *reading* the article it says...
It can now be revealed that the children taking part in the trial will be incentivised with the promise of up to £500 in vouchers for completing psychometric tests. These will measure the effect of the puberty blockers on their brains, including effects on impulse control and memory...Participants, aged under 16, will receive £30 vouchers for each of the 15 cognitive assessments they complete, as well as £15 vouchers for each of the three MRI scans they undertake, over the two years of the trial...Children in another arm of the trial, not taking the drugs, will be given £15 vouchers for each bone density scan and blood test they have.
Points to note
The children are already in the trial and have been randomised into one of the two arms, so they've already been consented.
Those in the intervention arm (on the drugs) are paid a nominal amount (£15 or £30) for each of the tests (cogntive assessments or MRI scans) they undertake
Those in the control arm (not on the drugs) are paid a nominal amount (£15) for each of the tests (bone density scan and blood test) they undertake
In short the participants are paid nominal sums to take part in tests they have already agreed to prior to assignment, and that this is an incentive to complete the trial, not an incentive to take part in the trial. Since a criticism of many studies in the trans sphere is dropout rate, this is a good idea.
There's a concept in RCT called "equipoise". The surgeons/doctors in a study must be genuinely undecided/unbiased as to which of the two arms is better, and the patients must be equally noncommittal: this enables randomisation to take place believably. In trans studies this is difficult since if the patients believe the study is biased or bent towards one outcome they will refuse to take part or simply bugger off to Europe and get the treatment they want there (Baroness Cass pointed this out as a reason to undertake the PATHWAY studies). The PATHWAY designers seem to have created a study that enables equipoise. The fact that these sums are paid to make sure they stay in the study prevents dropout
(Incidentally @isam, you said last night you wanted to know how tests of drugs in humans take place. You might want to google the term "CTIMP")
The “trial” is totally unethical by any conventional medical standard.
I don't entirely agree. There's a bit of me that thinks that Starmer has no political sense and so isn't up to the job of being PM, but there's also a bit of me that thinks that even though he's gotten into a right old mess there is an element of hysteria and opportunism in the criticism of Starmer. Over-all I'd say I'm a bit more sympathetic to him than I was. I think if he learns a lesson from this debacle he might improve his performance.
Right now even taking Starmer's weaknesses into account I can't see any better alternative PM from the Labour or opposition ranks. So if he's the best we've got I want him to do well as he can.
Dan Hodges is by far the worst.
Starmer can't hack it.. when criticised he went all what about this and what about that
I have no applications in my life where I need AI.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
Rare that we agree and I'm not saying I've never seen AI but it seems a case of using 200 words when 20 will do just fine.
It's a constipation of verbage designed literally to fill a space with nothing, words which I was once told didn't come easy but now do it would seem.
As for the world of work, weren't computers supposedly to allow us to have a life of leisure? Yes, it didn't for me either - indeed, the volume of information created by computer systems is analogous to the amount of verbage created by some AI products.
I warned my successors when I retired - you either manage the information or the information will manage you.
That's the end of the world of work, not through replacement but asphyxiation - we will drown in our information and suffocate on our words simultaenously.
In response to this morning's GDP figures, Chancellor Rachel Reeves says the UK the "fastest growing G7 economy in Europe" *. “The Government has the right economic plan to build a stronger and more secure economy, cutting the cost of living, cutting the national debt and creating the conditions for growth and investment in every part of the country.”
* The specificity kinda of gives the game away.
Cutting the cost of living.??? Its nonsense. Just shows how out of touch they are.
its gaslighting.
They have cut the cost of living - no more feeding your kids in the morning because the school will shovel some frosties down them instead - huzzah!
Anyone who says AI writes code for them is lying and does not work in software engineering.
I do and it does not. It has not improved at all in the last year, total stagnation.
Anyone doing serious work with these tools knows two things can be true at once: they’re astonishingly useful, and they still get things subtly wrong in ways that’ll bite you. But “no improvement at all” is just cope dressed up as confidence.
I’m extremely confident that AI is going to be a complete nothing, after using it professionally.
It’s not cope, it’s from my own experience. It’s overhyped rubbish.
When did you use it last and what tools were you using because I was equally as cynical as you up to christmas when I started a new project.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
Copilot managed to introduce a serious bug into Windows Notepad. That’s a pretty massive fail from me as the IT manager.
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
Microsoft sacked the majority of their QA team almost a decade ago. The users do the QA testing now.
Customer QA is not a great idea, eventually customers notice and change supplier.
Did anyone hear the ex civil servant who had a spat with Priti Patel a few yemrs back. He could have been Starmer speaking the Labour line. It was dreadful. Interview was on PM. Radio 4 this evening.
The Dutch has passed a law, as is being reported, that will tax unrealised capital gains at 36%. Due to come in 2028. Although there is some discussion as to changes.
I have no applications in my life where I need AI.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
I've had several friends and family (who are generally not technical people) mention they've resorted to using ChatGPT for one specific reason - medical advice. They either can't get a GP appointment in a reasonable timescale or the GP has brushed off their concerns, so they consult AI.
If my limited sample is in any way representative this is possibly quite widespread.
Did anyone hear the ex civil servant who had a spat with Priti Patel a few yemrs back. He could have been Starmer speaking the Labour line. It was dreadful. Interview was on PM. Radio 4 this evening.
Was that the one she should have been sacked over for bullying but successive PMs don't have the backbone to remove her, indeed she's been reappointed by gutless appeasers of bullying.
Of course she has very very powerful friends who cannot be upset.
I have no applications in my life where I need AI.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
Rare that we agree and I'm not saying I've never seen AI but it seems a case of using 200 words when 20 will do just fine.
It's a constipation of verbage designed literally to fill a space with nothing, words which I was once told didn't come easy but now do it would seem.
As for the world of work, weren't computers supposedly to allow us to have a life of leisure? Yes, it didn't for me either - indeed, the volume of information created by computer systems is analogous to the amount of verbage created by some AI products.
I warned my successors when I retired - you either manage the information or the information will manage you.
That's the end of the world of work, not through replacement but asphyxiation - we will drown in our information and suffocate on our words simultaenously.
“We will drown.” Maybe. Or maybe tooling shifts toward compression and synthesis rather than raw generation. Historically we oscillate — first we overproduce, then we build better filters.
I do think you’re right about the prolixity though. The cost of producing words has collapsed, and when production becomes trivial, volume explodes. That absolutely risks more noise, not more clarity.
The question isn’t whether more text gets generated — it will. The question is whether we get better at constraining, filtering and compressing it. Used lazily, these tools bloat everything. Used well, they actually reduce cognitive load.
In response to this morning's GDP figures, Chancellor Rachel Reeves says the UK the "fastest growing G7 economy in Europe" *. “The Government has the right economic plan to build a stronger and more secure economy, cutting the cost of living, cutting the national debt and creating the conditions for growth and investment in every part of the country.”
* The specificity kinda of gives the game away.
Cutting the cost of living.??? Its nonsense. Just shows how out of touch they are.
its gaslighting.
They have cut the cost of living - no more feeding your kids in the morning because the school will shovel some frosties down them instead - huzzah!
You may just but schools like the idea that children are fed before they start lessons and again at lunchtime - it means they are in a position to concentrate and are less likely to be disruptive.
Yes it may be useless parenting that means the schools are having to do it but it's a cheap solution to the problem...
Did anyone hear the ex civil servant who had a spat with Priti Patel a few yemrs back. He could have been Starmer speaking the Labour line. It was dreadful. Interview was on PM. Radio 4 this evening.
Was that the one she should have been sacked over for bullying but successive PMs don't have the backbone to remove her, indeed she's been reappointed by gutless appeasers of bullying.
Of course she has very very powerful friends who cannot be upset.
No it was the one where the gutless ex civil servant knew that the BBC wanted to hear but he refused to say it. Gutless civil servant portraying his colours imho.
I have no applications in my life where I need AI.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
Rare that we agree and I'm not saying I've never seen AI but it seems a case of using 200 words when 20 will do just fine.
It's a constipation of verbage designed literally to fill a space with nothing, words which I was once told didn't come easy but now do it would seem.
As for the world of work, weren't computers supposedly to allow us to have a life of leisure? Yes, it didn't for me either - indeed, the volume of information created by computer systems is analogous to the amount of verbage created by some AI products.
I warned my successors when I retired - you either manage the information or the information will manage you.
That's the end of the world of work, not through replacement but asphyxiation - we will drown in our information and suffocate on our words simultaenously.
Or you just tell AI to keep the answer as short as possible, ideally using bullet points...
Did anyone hear the ex civil servant who had a spat with Priti Patel a few yemrs back. He could have been Starmer speaking the Labour line. It was dreadful. Interview was on PM. Radio 4 this evening.
Was that the one she should have been sacked over for bullying but successive PMs don't have the backbone to remove her, indeed she's been reappointed by gutless appeasers of bullying.
Of course she has very very powerful friends who cannot be upset.
You.mean like Mandleson. And that's how he was reappointed because he knows where the bodies are buried.
The Dutch has passed a law, as is being reported, that will tax unrealised capital gains at 36%. Due to come in 2028. Although there is some discussion as to changes.
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
Not necessarily. Actually *reading* the article it says...
It can now be revealed that the children taking part in the trial will be incentivised with the promise of up to £500 in vouchers for completing psychometric tests. These will measure the effect of the puberty blockers on their brains, including effects on impulse control and memory...Participants, aged under 16, will receive £30 vouchers for each of the 15 cognitive assessments they complete, as well as £15 vouchers for each of the three MRI scans they undertake, over the two years of the trial...Children in another arm of the trial, not taking the drugs, will be given £15 vouchers for each bone density scan and blood test they have.
Points to note
The children are already in the trial and have been randomised into one of the two arms, so they've already been consented.
Those in the intervention arm (on the drugs) are paid a nominal amount (£15 or £30) for each of the tests (cogntive assessments or MRI scans) they undertake
Those in the control arm (not on the drugs) are paid a nominal amount (£15) for each of the tests (bone density scan and blood test) they undertake
In short the participants are paid nominal sums to take part in tests they have already agreed to prior to assignment, and that this is an incentive to complete the trial, not an incentive to take part in the trial. Since a criticism of many studies in the trans sphere is dropout rate, this is a good idea.
There's a concept in RCT called "equipoise". The surgeons/doctors in a study must be genuinely undecided/unbiased as to which of the two arms is better, and the patients must be equally noncommittal: this enables randomisation to take place believably. In trans studies this is difficult since if the patients believe the study is biased or bent towards one outcome they will refuse to take part or simply bugger off to Europe and get the treatment they want there (Baroness Cass pointed this out as a reason to undertake the PATHWAY studies). The PATHWAY designers seem to have created a study that enables equipoise. The fact that these sums are paid to make sure they stay in the study prevents dropout
(Incidentally @isam, you said last night you wanted to know how tests of drugs in humans take place. You might want to google the term "CTIMP")
The “trial” is totally unethical by any conventional medical standard.
On the contrary, it is totally unethical to do such trials. How else is medicine to progress? Anecdote?
Obviously any trial on children requires careful thought about issues of consent.
this is something from long before the trench of Mandy emails, from a sense he was blocking Civil Service reform. Kemi ordered for him to be kept in post, but could he not become even more free to speak out once not on gardening leave but signed for private industry? So the opposite of a cover up unless we’re spinning it?
The timing is very interesting though and worth the media probing and PMQ questions probing.
I think it’s all pointing towards my analysis I have been explaining, but no one listening again - the establishment, civil service and security services wanted Mandleson to get the gig.
I have no applications in my life where I need AI.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
Rare that we agree and I'm not saying I've never seen AI but it seems a case of using 200 words when 20 will do just fine.
It's a constipation of verbage designed literally to fill a space with nothing, words which I was once told didn't come easy but now do it would seem.
As for the world of work, weren't computers supposedly to allow us to have a life of leisure? Yes, it didn't for me either - indeed, the volume of information created by computer systems is analogous to the amount of verbage created by some AI products.
I warned my successors when I retired - you either manage the information or the information will manage you.
That's the end of the world of work, not through replacement but asphyxiation - we will drown in our information and suffocate on our words simultaenously.
Or you just tell AI to keep the answer as short as possible, ideally using bullet points...
Modern communication:
Person A tells the AI to expand their bullet points into a long email.
Person B tells the AI to summarise the long email into bullet points.
Comments
Hey, Keir !!
I spent a lot of time over the weekend writing code with Claude. And it was very clear that we will never ever write code by hand again. It doesn't make any sense to do so.
https://x.com/adityaag/status/2018496292608155756?s=20
But Incorrect Horse Battery says Claude Code is rubbish. I think it is probably more you aren't using them very well. Are they 100% correct, no, but humans makes masses of mistakes. Is Claude Code + Opus absolute elite tier coder, yes.
It created a bug I’ve been fixing literally today
One of our juniors is obsessed with it.
I write mostly Verilog and assembly language and design much of the hardware that hosts my code. I do not expect to be replaced by an LLM any time in the foreseeable future. But if I spent my days writing Python or maintaining some shonky mobile app, I'd be worried.
You have your experiences, I have mine. But allowing Claude into our organisation has been a really, really bad move and I can only speak to that.
It will not replace a single job here.
Because that's the important bit here - Co-Pilot is still a pile of horseshite and nothing is going to fix it but some of the other tools are getting rather useful especially as their tooling improves.
Now is it perfect, nope, but the output is more usable than most of the Indians devs I deal with on a day to day basis
The secret is planning. You spend time chatting through the plan. As long as you do that, and you go step by step through what the plan is, you don't just do "code me this", it is super human level coder. Its like have a team of really really good devs.
That's asking for trouble...
So you merged AI-generated code into something important without fully understanding it, without adequate tests, and without catching the bug in review… and the conclusion is “the tool is rubbish”.
That’s like blaming Visual Studio because you copy-pasted from Stack Overflow and prod fell over.
The obvious error isn’t in Claude. It’s in the workflow.
If a junior is “obsessed with it” and generating bugs, that’s not an AI problem — that’s a supervision and engineering standards problem. You don’t let juniors ship unchecked code whether it came from their own brain or a stochastic parrot.
https://www.telegraph.co.uk/gift/af4cc570a8281291
The Telegraph has revealed that children, including ones with learning difficulties, have been promised up to £500 in vouchers, which can be redeemed at retailers like XBOX & Uber Eats, if they take part in an NHS 'transgender' trial to block their puberty
So I’m now just reviewing his code particularly carefully and it’s taking at least twice as long as if he’d just written it from scratch.
I can read code now and tell when AI wrote it.
https://x.com/coldwarsteve/status/2021984067852312679?s=61&t=LYVEHh2mqFy1oUJAdCfe-Q
Windows 11 has genuinely come close to being unusable over the past few months, as MS have their AI writing the code and seemingly no-one doing the QA on it.
I used it out of interest for some basic analytics of a spreadsheet I’d made and it added two numbers incorrectly.
Its the enabling of the context in terminal / IDE plus some secret sauce that Cluade Opus currently has that makes it stand out. I hear things Kimi are catching up though.
To be fair, I was already irritated because at 3am I was woken by by a text from my bank telling me that they'd been suspicious of, and therefore blocked, a payment and sorting it out when I was finally properly conscious took quite a long time. Particularly because said payment was a scam of some sort. Anyone heard of AQ*MAIDS? I haven't.
So that's my credit card out of action for three or so days!
Like when it told me multiple times that Angular does not support @if syntax.
Nobody doing anything serious with AI uses copilot,
I'm using Cursor with a very carefully crafted set of rules, domain knowledge and memories.
Some of the most elite coders in the world are saying these tools, especially Claude Code is now top tier. But you seem to have a terrible experience e.g. are you just using free plans via web interface or are you using the max tier models with CLI integration?
No wonder Lisa has referred Telegraph Mail merger.
Best course, shut em both down like the NOTW
Slight problem it's been in production for years and a lot of businesses use it for Project management and Field Service organisation..
It's definitely the case that Microsoft's quality control (which was never that great) is getting far far worse.
Now same people are moaning as he gets the boot.
Our company pays for it because they’re convinced it’s the future (another reason I think it’s overhyped).
So she's clearly upset some people along the way.
Fine line between being authoratative and being a bully.
Not old guard
Doubt she knew Epstein
Give her a chance.
Many in right wing media she didn't get the job in the first place now samd cabal treat her as if she's Cruella D..
Media has changed not her.
Give a a proper chance.
But you do you.
https://www.bbc.co.uk/news/articles/c5yqr8eggvwo
- Very senior people think it's amazing. Because they don't do much actual work, they delegate. This gives them plausible answers much quicker than asking someone, and certainly more so than if they tried to do the work themselves.
- Very junior people see AI as a shortcut where their knowledge or skills are lacking. It probably improves their output speed and quality. My fear is it also slows their learning on how to be better.
- Experienced, non-management staff find AI offers much more marginal gains. Because they are already very efficient at completing tasks, can send emails or do calculations or make presentations quickly and to a high quality without needing AI. AI helps, but its assistance is more marginal than people in the senior or junior group think it is.
The challenge is going to be how to train the experienced workers of the future if juniors are always taking short cuts via AI that does not yet exceed the abilities of experienced people... And how to stop management obsessing over it when its usefulness is limited.
One issue I can foresee is companies cutting back on juniors (or expect them to churn lots more out), especially in things like coding, then in 10 years going shit we don't have any experienced staff because the ones we had are retiring and we didn't hire enough / they weren't trained up.
The only point I would disagree with is not that helpful for experienced. I am experience and it has made me massively more productive, but I can read / understand the technical stuff, make the plans with the LLM and check that it is going to do roughly the right thing, then can read the code quickly and understand what it does to weed out the mistakes. Also, I know what to ask it to check for to eliminate common mistakes. Along with the planning, the asking for LLM output to check themselves in the correct way is another crucial step.
Pile of shite.
Where you do have a point is the language you choose does matter in as much as the LLMs are trained on existing repos, the more common the language the "better" the training.
One big thing that Claude Code gives you and I have found the really mindblowing, I have seen it numerous times say to itself, I don't know or this is a new version that has new functionality, just go and download all the docs / tutorials / blogs, read them, then can use that new functionality.
Now if you want to tell me some code in Fortran or Ada to runs some nuclear power plant and 5 people ever seen that codebase, probably, it does need the background data knowledge.
A Ryanair copilot is earning around £50 per flight hour, on a zero-hours contract, and has a £30k debt for their type rating that they need to pay off.
There’s caravan parks and car parks near Stanstead filled with FR first officers living in trailers and vans.
They will continue to be.
Some will have to adapt. I would not want to be getting paid to write nonsense when AI can just generate it for instance.
I can see why @leon is worried.
Good writers borrow, great writers steal, as I've just stolen.
I am sure down the line new jobs come along but that isn't comforting for those in the firing line in 5 years time. Also, its was the West's advantage, this more educated workforce to do service jobs. That advantage is being swept away e.g. translation, you don't need it for general day to day things now. For Medical / Legal scenarios yes, but not for just contacting other people inquiring about things.
It can now be revealed that the children taking part in the trial will be incentivised with the promise of up to £500 in vouchers for completing psychometric tests. These will measure the effect of the puberty blockers on their brains, including effects on impulse control and memory...Participants, aged under 16, will receive £30 vouchers for each of the 15 cognitive assessments they complete, as well as £15 vouchers for each of the three MRI scans they undertake, over the two years of the trial...Children in another arm of the trial, not taking the drugs, will be given £15 vouchers for each bone density scan and blood test they have.
Points to note
- The children are already in the trial and have been randomised into one of the two arms, so they've already been consented.
- Those in the intervention arm (on the drugs) are paid a nominal amount (£15 or £30) for each of the tests (cogntive assessments or MRI scans) they undertake
- Those in the control arm (not on the drugs) are paid a nominal amount (£15) for each of the tests (bone density scan and blood test) they undertake
In short the participants are paid nominal sums to take part in tests they have already agreed to prior to assignment, and that this is an incentive to complete the trial, not an incentive to take part in the trial. Since a criticism of many studies in the trans sphere is dropout rate, this is a good idea.There's a concept in RCT called "equipoise". The surgeons/doctors in a study must be genuinely undecided/unbiased as to which of the two arms is better, and the patients must be equally noncommittal: this enables randomisation to take place believably. In trans studies this is difficult since if the patients believe the study is biased or bent towards one outcome they will refuse to take part or simply bugger off to Europe and get the treatment they want there (Baroness Cass pointed this out as a reason to undertake the PATHWAY studies). The PATHWAY designers seem to have created a study that enables equipoise. The fact that these sums are paid to make sure they stay in the study prevents dropout
(Incidentally @isam, you said last night you wanted to know how tests of drugs in humans take place. You might want to google the term "CTIMP")
It's a particular disaster for Russia, where production is being negatively impacted by lack of access to Western production equipment.
https://x.com/athlete365/status/2021941246449922445
The Tory Cocoon Club forget all the shithousery during Covid and the aftermath
Utter shithousery
Wormold should thank his lucky stars.
The current LOTO in the days before they took Twitter off her, famously said 10% of all Civil Servants should be on prison for being so bad
Jibes aside (apologies Leon) it is really quite challenging to work out quite which jobs are likely to succumb to AI. If you'd asked me some years ago I wouldn't particularly have said developers for example, and yet in hindsight its obvious in that all they do is write pretty much the same thing as other developers but with variations.
I don't know, but I do wonder, who cleans Waymo taxis?
AGI is a long way off, and none of the current 'AIs' can (in my view) ever result in such a thing. They're all on a very interesting path, but regurgitation isn't the way.
They should not be remaining in a trial they would otherwise drop out of, solely due to fiscal considerations.
Still less, one that I would pay for.
I suspect that I am far, far more reflective of AI than the tech bros who own it.
Starmer is weak... weak....weak....
https://x.com/zoryalondonsk/status/2021983227205104067
It's a constipation of verbage designed literally to fill a space with nothing, words which I was once told didn't come easy but now do it would seem.
As for the world of work, weren't computers supposedly to allow us to have a life of leisure? Yes, it didn't for me either - indeed, the volume of information created by computer systems is analogous to the amount of verbage created by some AI products.
I warned my successors when I retired - you either manage the information or the information will manage you.
That's the end of the world of work, not through replacement but asphyxiation - we will drown in our information and suffocate on our words simultaenously.
Sounds nuts.
https://x.com/bitcoinnewscom/status/2021978390870347923?s=61
If my limited sample is in any way representative this is possibly quite widespread.
I saw the house voted against Canada tarrifs.
Of course she has very very powerful friends who cannot be upset.
I do think you’re right about the prolixity though. The cost of producing words has collapsed, and when production becomes trivial, volume explodes. That absolutely risks more noise, not more clarity.
The question isn’t whether more text gets generated — it will. The question is whether we get better at constraining, filtering and compressing it. Used lazily, these tools bloat everything. Used well, they actually reduce cognitive load.
Yes it may be useless parenting that means the schools are having to do it but it's a cheap solution to the problem...
Obviously any trial on children requires careful thought about issues of consent.
Kemi ordered for him to be kept in post, but could he not become even more free to speak out once not on gardening leave but signed for private industry? So the opposite of a cover up unless we’re spinning it?
The timing is very interesting though and worth the media probing and PMQ questions probing.
I think it’s all pointing towards my analysis I have been explaining, but no one listening again - the establishment, civil service and security services wanted Mandleson to get the gig.
https://news.sky.com/story/sir-jim-ratcliffe-scolds-tories-over-handling-of-economy-and-immigration-after-brexit-13139088
"I mean, no small island like the UK could cope with vast numbers of people coming into the UK.
"I mean, it just overburdens the National Health Service, the traffic service, the police, everybody.
"The country was designed for 55 or 60 million people and we've got 70 million people and all the services break down as a consequence.
Person A tells the AI to expand their bullet points into a long email.
Person B tells the AI to summarise the long email into bullet points.