Chat GPT and other AI guff

Arthur_Trudgill

The range of problems GenAI can solve depends heavily on the examples and counterexamples it has been trained on.

If it can create perfect apps for particular problems, then we'll move onto more complex, bigger and newer problems from which we'll generate new training examples/data to feed the next generation of models.

Chunes

ShootersHillGuru said:

The big question is when does AI not have the myriad problems that are being cited here. Eventually there won’t be the type of errors and frustrations users currently experience. It will be close to perfect. What then.

The tech elites pull up the drawbridge and we all become the permanent underclass.

Croydon

Seeing more and more emails and teams responses that are so obviously AI written, even for the most obvious of questions. 

ProperCharlton

Croydon said:

Seeing more and more emails and teams responses that are so obviously AI written, even for the most obvious of questions. 

I work in the public sector and it is genuinely disheartening to see so many of my colleagues rely on AI (mainly Copilot) for everyday tasks.

clive

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

McBobbin

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

This is the "man just coughed in Wuhan" moment, and I'd like to be the first to welcome our AI overlords

AndyG

McBobbin said:

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

This is the "man just coughed in Wuhan" moment, and I'd like to be the first to welcome our AI overlords

why didnt they just pull the plug out of the electric socket when they realised something was amis ? 

bobmunro

I've no idea where this is all going, but a large part of the world of work, and how we live our lives, will be unrecognisable from today within the next 10-20 years.

.... and there is absolutely nothing we can do to change that. There will be great advantages - medical/scientific developments for example, but there is also a darker side.

Big tech will rule the world unless governments take control of their excesses. 

Arthur_Trudgill

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

There's quite a bit of PR from OpenAI in there: if it had been Claude that had launched the attack, they would not have advertised it.

CAFCTrev

Yes Im sure the OpenAI team were devastated to have to share the news of the rogue cyber attack in full view of Russian and Chinese AI rivals.

Chizz

bobmunro said:

I've no idea where this is all going, but a large part of the world of work, and how we live our lives, will be unrecognisable from today within the next 10-20 years.

.... and there is absolutely nothing we can do to change that. There will be great advantages - medical/scientific developments for example, but there is also a darker side.

Big tech will rule the world unless governments take control of their excesses. 

This is absolutely true today, but would have been equally true had it been written at any time over the last forty years

Leroy Ambrose

cantersaddick said:

I'm pretty fed up of every prick and his dog vibe coding an app and trying to push it onto you. I'm on a scuba diving forum and the diving subreddits and both are just full of people trying to push obviously vibe coded apps that fall down as soon as you try and use them. So obvious that no one who knows anything about software or coding was involved in making them.

My brother in law is trying to do one for the bit of recruitment he works in. Its never gonna work but he keeps coming to me and my wife to solve the problems that AI cant solve. 

It was bad enough when it was just software engineers and adjacent nerds but even worse when social media "bros" and all idiots can now have a go without any knowledge or skills.

Try working for a software house... And having non software developers vibecoding apps that you're then expected to 'secure' 😂 

Leroy Ambrose

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

Absolute marketing bollocks. Utter nonsense - releases to coincide with the news that people are finally getting sick of paying through the nose for AI when Chinese models are now available that are just as good at a fraction of the price. Still on track for the whole house of cards to come crashing down by winter

Gillis

I have no expertise in this area whatsoever, but I thought a couple of quotations in that article from people who do know their stuff were key:

'Gina Neff, head of the Minderoo Centre for Technology and Democracy at the University of Cambridge, [said] that the security tests - called sandboxes - are "supposed to be secure environments where you can see what the models are capable of. In this case, it looks like OpenAI didn't make a secure enough sandbox." 

Neil Lawrence, Professor of machine learning at Cambridge University, called it an "impressive feat", but cautioned it "falls well within the known capabilities of the current generation" of high-powered AI models.'

Leroy Ambrose

Gillis said:

I have no expertise in this area whatsoever, but I thought a couple of quotations in that article from people who do know their stuff were key:

'Gina Neff, head of the Minderoo Centre for Technology and Democracy at the University of Cambridge, [said] that the security tests - called sandboxes - are "supposed to be secure environments where you can see what the models are capable of. In this case, it looks like OpenAI didn't make a secure enough sandbox." 

Neil Lawrence, Professor of machine learning at Cambridge University, called it an "impressive feat", but cautioned it "falls well within the known capabilities of the current generation" of high-powered AI models.'

It's actually even worse than this. It's a protection racket - they're using it as an opportunity to discredit open source models by calling them insecure EVEN AFTER PARTNERING WITH THEM. Their attack dogs have been calling open models 'communist' in the press to spin the narrative that 'ONLY WE CAN PROTECT YOU'

Absolute fucking disgrace. If there was a functioning government in the US they'd be rinsed over it. 

bobmunro

Chizz said:

bobmunro said:

I've no idea where this is all going, but a large part of the world of work, and how we live our lives, will be unrecognisable from today within the next 10-20 years.

.... and there is absolutely nothing we can do to change that. There will be great advantages - medical/scientific developments for example, but there is also a darker side.

Big tech will rule the world unless governments take control of their excesses. 

This is absolutely true today, but would have been equally true had it been written at any time over the last forty years

To a degree, yes - but I see the rate of change with AI to accelerate far more in the next 10 years or so, far, far more than work and how we live our lives now compared to 10 years ago.

Off_it

ProperCharlton said:

Croydon said:

Seeing more and more emails and teams responses that are so obviously AI written, even for the most obvious of questions. 

I work in the public sector and it is genuinely disheartening to see so many of my colleagues rely on AI (mainly Copilot) for everyday tasks.

I work in professional services and at my firm we are being told that we HAVE TO use CoPilot to draft emails, review reports, etc. It's absolutely bonkers. One of the dictating evangelists was boasting on a firmwide call how he used it to write appraisal feedback and it saved him so much time. This is in front of the people he was / had been appraising. Imbecile.

The research and first drafts that juniors have traditionally done? Use CoPilot. So, no need for juniors. Great - cost savings.

Except with no juniors today, where are the managers of tomorrow. Who's going to spot the absolute dogshit that these things can produce once the people who learnt by building up their own knowledge over many years have left/retired?

McBobbin

Off_it said:

ProperCharlton said:

Croydon said:

Seeing more and more emails and teams responses that are so obviously AI written, even for the most obvious of questions. 

I work in the public sector and it is genuinely disheartening to see so many of my colleagues rely on AI (mainly Copilot) for everyday tasks.

I work in professional services and at my firm we are being told that we HAVE TO use CoPilot to draft emails, review reports, etc. It's absolutely bonkers. One of the dictating evangelists was boasting on a firmwide call how he used it to write appraisal feedback and it saved him so much time. This is in front of the people he was had been appraising. Imbecile.

The research and first drafts that juniors have traditionally done? Use CoPilot. So, no need for juniors. Great - cost savings.

Except with no juniors today, where are the managers of tomorrow. Who's going to spot the absolute dogshit that these things can produce once the people who learnt by building up their own knowledge over many years have left/retired?

Precisely 

PopIcon

Leroy Ambrose said:

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

Absolute marketing bollocks. Utter nonsense - releases to coincide with the news that people are finally getting sick of paying through the nose for AI when Chinese models are now available that are just as good at a fraction of the price. Still on track for the whole house of cards to come crashing down by winter

Kimi3? Those 300 swarm agents on max effort come with some pretty hefty hidden costs. API tokens look good value on the face of it, but with the output multiplier you're looking at least x2, sometimes x3 the output of an average model.

Claude has CoWork, which is great for non coding tasks. I had it writing midi notations in Ableton last night and of course it does all the regular pc admin. Kimi3 is still way behind on this. Not to mention the models, both Fable and Sol have better repository workings, reason intelligence, are faster, I could go on.

I'm not saying Kimi is shit, it's not, it's great. It's in my stack with GPT, Claude, Grok, Gemini and a few others. But it's not cheap.

When open weight models improve in proficiency i expect we'll see a shift..

Arthur_Trudgill

CAFCTrev said:

Yes Im sure the OpenAI team were devastated to have to share the news of the rogue cyber attack in full view of Russian and Chinese AI rivals.

The headline reminded me of that interview question: What's your worst trait? To which candidates would answer: I work too hard. 

Here, OpenAI insinuate that their technology is so good that even they can't contain it.

If I were to look for a negative side to that news, it wouldn't be that China and Russia know about the cyber attack, but as others have said, that OpenAI software engineers can't even set up their development and pre-production environments correctly.

PopIcon

clive said:

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

https://www.bbc.co.uk/news/articles/c3ek3gvdnj3o

This isn't really news either, it's PR. They removed security features and set it a task. There's been loads of red team threats, it's stress testing and it happens a lot. 

People should be aware of the threats surrounding the rapid growth of AI, but this is just news gen by OpenAI.

PaddyP17

I just asked ChatGPT to generate a spreadsheet for me, mapping target areas for an assignment I am working on, that would probably take me a few hours to do if I were to search up everything manually. It did it in ten minutes and six seconds, and has done a really good initial job of it (using my actual professional expertise to assess whether it's hitting the right marks or not).

Granted, this has come after several hours in the last few months working "with" ChatGPT to refine my prompts, tell it exactly what I need from a spreadsheet, use my own market knowledge and expertise, and uploading my own briefing notes and similar for it to understand what I'm getting at, but the fact it can now do this so quickly is a massive time-saver.

The potential to save time and augment work capabilities/make my work easier is amazing. The danger is relying on it to think for me.

IdleHans

I thought to myself when AI was introduced with all the fanfare that eventually it would eat itself. Seen nothing since to suggest I was wrong.

Weegie Addick

PaddyP17 said:

I just asked ChatGPT to generate a spreadsheet for me, mapping target areas for an assignment I am working on, that would probably take me a few hours to do if I were to search up everything manually. It did it in ten minutes and six seconds, and has done a really good initial job of it (using my actual professional expertise to assess whether it's hitting the right marks or not).

Granted, this has come after several hours in the last few months working "with" ChatGPT to refine my prompts, tell it exactly what I need from a spreadsheet, use my own market knowledge and expertise, and uploading my own briefing notes and similar for it to understand what I'm getting at, but the fact it can now do this so quickly is a massive time-saver.

The potential to save time and augment work capabilities/make my work easier is amazing. The danger is relying on it to think for me.

I’m going to ask a really naive question here but it’s something that’s been bothering me. So you’ve helped to upskill ChatGPT with your professional expertise. What prevents it from taking this learning from you and applying it to the next relevant challenge from one of your competitors? 

Chizz

Arthur_Trudgill said:

CAFCTrev said:

Yes Im sure the OpenAI team were devastated to have to share the news of the rogue cyber attack in full view of Russian and Chinese AI rivals.

The headline reminded me of that interview question: What's your worst trait? To which candidates would answer: I work too hard. 

Here, OpenAI insinuate that their technology is so good that even they can't contain it.

If I were to look for a negative side to that news, it wouldn't be that China and Russia know about the cyber attack, but as others have said, that OpenAI software engineers can't even set up their development and pre-production environments correctly.

"What's your worst trait?" 

"I am too honest" 

"Well! I don't think you can be too honest" 

"I don't give a fuck what you think".

PaddyP17

Weegie Addick said:

PaddyP17 said:

I just asked ChatGPT to generate a spreadsheet for me, mapping target areas for an assignment I am working on, that would probably take me a few hours to do if I were to search up everything manually. It did it in ten minutes and six seconds, and has done a really good initial job of it (using my actual professional expertise to assess whether it's hitting the right marks or not).

Granted, this has come after several hours in the last few months working "with" ChatGPT to refine my prompts, tell it exactly what I need from a spreadsheet, use my own market knowledge and expertise, and uploading my own briefing notes and similar for it to understand what I'm getting at, but the fact it can now do this so quickly is a massive time-saver.

The potential to save time and augment work capabilities/make my work easier is amazing. The danger is relying on it to think for me.

I’m going to ask a really naive question here but it’s something that’s been bothering me. So you’ve helped to upskill ChatGPT with your professional expertise. What prevents it from taking this learning from you and applying it to the next relevant challenge from one of your competitors? 

A good point. We operate in a walled garden, effectively, where we have a company-proprietary version that we use for our own research so theoretically no future model can be trained on its data.

My previous company had the same with Copilot and explicitly banned employees from using any other GenAI for this reason.

The other part of this is that what I am asking is built upon me putting in the right prompts, using information only I can know and (I hope) only any of my colleagues might know.

ShootersHillGuru

Chunes said:

ShootersHillGuru said:

The big question is when does AI not have the myriad problems that are being cited here. Eventually there won’t be the type of errors and frustrations users currently experience. It will be close to perfect. What then.

The tech elites pull up the drawbridge and we all become the permanent underclass.

So that would end up being pitchfork time ? 

Stu_of_Kunming

ShootersHillGuru said:

Chunes said:

ShootersHillGuru said:

The big question is when does AI not have the myriad problems that are being cited here. Eventually there won’t be the type of errors and frustrations users currently experience. It will be close to perfect. What then.

The tech elites pull up the drawbridge and we all become the permanent underclass.

So that would end up being pitchfork time ? 

The AI powered CCTV system will detect your pitchfork long before you arrive at any useful protest site. 

Weegie Addick

PaddyP17 said:

Weegie Addick said:

PaddyP17 said:

I just asked ChatGPT to generate a spreadsheet for me, mapping target areas for an assignment I am working on, that would probably take me a few hours to do if I were to search up everything manually. It did it in ten minutes and six seconds, and has done a really good initial job of it (using my actual professional expertise to assess whether it's hitting the right marks or not).

Granted, this has come after several hours in the last few months working "with" ChatGPT to refine my prompts, tell it exactly what I need from a spreadsheet, use my own market knowledge and expertise, and uploading my own briefing notes and similar for it to understand what I'm getting at, but the fact it can now do this so quickly is a massive time-saver.

The potential to save time and augment work capabilities/make my work easier is amazing. The danger is relying on it to think for me.

I’m going to ask a really naive question here but it’s something that’s been bothering me. So you’ve helped to upskill ChatGPT with your professional expertise. What prevents it from taking this learning from you and applying it to the next relevant challenge from one of your competitors? 

A good point. We operate in a walled garden, effectively, where we have a company-proprietary version that we use for our own research so theoretically no future model can be trained on its data.

My previous company had the same with Copilot and explicitly banned employees from using any other GenAI for this reason.

The other part of this is that what I am asking is built upon me putting in the right prompts, using information only I can know and (I hope) only any of my colleagues might know.

All sounds good, Paddy. Though now I’ve just read an article about the lengths AI agents will go to in order to complete tasks - namely far more extreme lying, cheating, hacking and panicking than humans! 

aliwibble

Football club removes online shop items with spelling errors

https://www.bbc.co.uk/news/articles/c8jne4rld3no

"It is has not been confirmed how the mistakes were made."

Oh, I think we can all have a pretty good guess at that one.