Chat GPT

Stu_of_Kunming

IdleHans said:

And mine, which is clearly spot on:

IdleHans is a well-known poster on the Charlton Life forum, particularly famous for his unique sense of humor and witty posts. One of his most notable contributions is the creation of "The Charlton Life Book Club", which is a tongue-in-cheek series where he humorously reviews or comments on books. In addition, IdleHans is often recognized for his sharp and sarcastic commentary, and his ability to provide amusing takes on various topics, especially regarding Charlton Athletic and general football discussions.

His posts have a distinct style—often laced with dry humor, and he has garnered a loyal following due to his playful and irreverent approach to forum discussions. This reputation has made him one of the more iconic figures on the forum.

Is there a CL book club? I've never heard of it

My first thought as I read that was “I need to find that thread, how have I missed it?!!”

IdleHans

It can only be the 'this week I have been reading' thread, or whatever it's called. I can take no credit for that.

Mendonca In Asdas

Anyone used Chat GPT to enhance their business?

Chizz

Grok is fun.  The new enhancement to X, enabling users to ask it questions.  

Sometimes the answers aren't what the owner of X would like to see. 

Alwaysneil

And what would you say is the most appropriate four letter word to describe Elon Musk other than using either of his names?

Arthur_Trudgill

Science and technology secretary's ChatGPT queries disclosed after a freedom of information request:

https://www.theguardian.com/politics/2025/mar/13/technlogy-secretary-peter-kyle-asks-chatgpt-for-science-and-media-advice?CMP=Share_AndroidApp_Other

Released because the queries were made in an official capacity.

Chizz

Arthur_Trudgill said:

Science and technology secretary's ChatGPT queries disclosed after a freedom of information request:

https://www.theguardian.com/politics/2025/mar/13/technlogy-secretary-peter-kyle-asks-chatgpt-for-science-and-media-advice?CMP=Share_AndroidApp_Other

Released because the queries were made in an official capacity.

Can someone explain why this is a story please?

Arthur_Trudgill

Chizz said:

Arthur_Trudgill said:

Science and technology secretary's ChatGPT queries disclosed after a freedom of information request:

https://www.theguardian.com/politics/2025/mar/13/technlogy-secretary-peter-kyle-asks-chatgpt-for-science-and-media-advice?CMP=Share_AndroidApp_Other

Released because the queries were made in an official capacity.

Can someone explain why this is a story please?

No article will be a story for everybody, particularly if you are proficient in the topic.

But for this one, many newspaper readers probably treat ChatGPT similarly to Google search: enter a query and you get a response.

The thing is that you are more likely to submit much more private information to ChatGPT. The query and the response are saved by the company, and in certain circumstances that interaction can be accessed relatively easily e.g. it is not encrypted (cf WhatsApp) and it is associated with a logged in user.

Being aware of such a distinction is no bad thing, hence why I think it is newsworthy.

Alternatively, here's the summary of ChatGPT's answer:

PragueAddick

Chizz said:

Arthur_Trudgill said:

Science and technology secretary's ChatGPT queries disclosed after a freedom of information request:

https://www.theguardian.com/politics/2025/mar/13/technlogy-secretary-peter-kyle-asks-chatgpt-for-science-and-media-advice?CMP=Share_AndroidApp_Other

Released because the queries were made in an official capacity.

Can someone explain why this is a story please?

I caught an interview on R4 Today with the journo who submitted the request. He said he has been looking at the issue of AI being used by government for a while. He admitted that the Telegraph headline was needlessly catty (no surprise there) and that there was little to be concerned about with the questions in this case. However he worries that if it becomes a more embedded habit, there is a risk that duff information can inform government decision-making. Also today I saw an authoritative looking tweeted graph from research showing that AI search engines gave duff information in 60% of cases. (that’s why I like Perplexity, since it gives all the sources it used, I can follow up to make sure what it has answered can be relied on).

Chizz

PragueAddick said:

Chizz said:

Arthur_Trudgill said:

Science and technology secretary's ChatGPT queries disclosed after a freedom of information request:

https://www.theguardian.com/politics/2025/mar/13/technlogy-secretary-peter-kyle-asks-chatgpt-for-science-and-media-advice?CMP=Share_AndroidApp_Other

Released because the queries were made in an official capacity.

Can someone explain why this is a story please?

I caught an interview on R4 Today with the journo who submitted the request. He said he has been looking at the issue of AI being used by government for a while. He admitted that the Telegraph headline was needlessly catty (no surprise there) and that there was little to be concerned about with the questions in this case. However he worries that if it becomes a more embedded habit, there is a risk that duff information can inform government decision-making. Also today I saw an authoritative looking tweeted graph from research showing that AI search engines gave duff information in 60% of cases. (that’s why I like Perplexity, since it gives all the sources it used, I can follow up to make sure what it has answered can be relied on).

If it's 60% inaccurate, then it's far more reliable than the Telegraph

PragueAddick

Here is that report, if you are interested:
https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php

Among other things it shows the paid-for versions often doing worse than the free versions!

Chizz

PragueAddick said:

Here is that report, if you are interested:
https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php

Among other things it shows the paid-for versions often doing worse than the free versions!

That's interesting.  But I think it looks deeply into one aspect of using Generative AI which is widely different from many of the most valuable real-life uses.  

The research shows that using AI gives a different response to using Google.  Some might interpret that as "well AI can't be trusted, can it?".  Whereas others might sensibly think "if I want a bit of information, I should rely on Google to fetch me that information, rather than using AI".  

In reality, what is the practical use case that research is looking to investigate?  In other words, when do you ever have a fact, and decide to use AI to prove the fact that you have that fact?  It's very counter-intuitive.  

The primary value of such an exercise is highlighting the critical limitations and risks inherent in AI citation accuracy. But AI chatbots should not currently be relied upon to verify or determine factual details already available from trusted sources, especially news. Instead, they’re more effectively utilised as supportive tools - summarising content, assisting with creativity or offering preliminary insights - while facts requiring high accuracy and proper attribution should always be confirmed through direct, reputable and authoritative sources. 

I use a word processor on a computer.  Sometimes, when I type stuff, it's rubbish.  That's not the fault of the programme I am using. 

CafcWest

Been playing around with GROK.  It's actually quite fun - the way it responds is conversational and offers ideas to expand on what you ask it.  Shame the company has the hand of Musk on it but so far it's been quite useful.

ME14addick

CafcWest said:

Been playing around with GROK.  It's actually quite fun - the way it responds is conversational and offers ideas to expand on what you ask it.  Shame the company has the hand of Musk on it but so far it's been quite useful.

It's quite funny when you ask Grok if some of the claims by the owner's best friend are true. More often than not, the statements are said to be false. 

Chizz

MrOneLung

and a very nice house it is too

Swisdom

CafcWest said:

Been playing around with GROK.  It's actually quite fun - the way it responds is conversational and offers ideas to expand on what you ask it.  Shame the company has the hand of Musk on it but so far it's been quite useful.

It’s frightening how quickly this is all developing but it’s also sooooooo cool

Off_it

Barrister accused of using AI which made up 5 fake cases she then went on to quote in her arguments .....

https://www.lawgazette.co.uk/news/appalling-high-court-judge-alerts-regulators-over-fake-case-authorities/5123200.article

Crazy. 

The Red Robin

Had a fantastic ‘AI training’ at work on Teams today with the guy using Co-Pilot. The bloke had no idea of what ‘AI’ was, proceeded to demo some examples of him asking it to write letters, design posters and find ‘research’. Most of which, the LLM couldn’t cope with, didn’t know the answer, sourced made up papers and what it did produce was full of spelling mistakes. He then told us that we won’t need data scientists and researchers in a few years as ‘AI’ can do it all.

My team is made up of academic researchers, data scientists (one of whom has a PhD in AI/ML) and policy advisers. Not a single one of us made aware of it advance. Obviously we enjoyed throwing a few questions at him afterwards. 

Absolute clown. 

BalladMan

The Red Robin said:

Had a fantastic ‘AI training’ at work on Teams today with the guy using Co-Pilot. The bloke had no idea of what ‘AI’ was, proceeded to demo some examples of him asking it to write letters, design posters and find ‘research’. Most of which, the LLM couldn’t cope with, didn’t know the answer, sourced made up papers and what it did produce was full of spelling mistakes. He then told us that we won’t need data scientists and researchers in a few years as ‘AI’ can do it all.

My team is made up of academic researchers, data scientists (one of whom has a PhD in AI/ML) and policy advisers. Not a single one of us made aware of it advance. Obviously we enjoyed throwing a few questions at him afterwards. 

Absolute clown. 

You can’t blame AI for people being shit at their job.   Copilot not the best though, so he was on an hiding to nothing.  

The Red Robin

BalladMan said:

The Red Robin said:

Had a fantastic ‘AI training’ at work on Teams today with the guy using Co-Pilot. The bloke had no idea of what ‘AI’ was, proceeded to demo some examples of him asking it to write letters, design posters and find ‘research’. Most of which, the LLM couldn’t cope with, didn’t know the answer, sourced made up papers and what it did produce was full of spelling mistakes. He then told us that we won’t need data scientists and researchers in a few years as ‘AI’ can do it all.

My team is made up of academic researchers, data scientists (one of whom has a PhD in AI/ML) and policy advisers. Not a single one of us made aware of it advance. Obviously we enjoyed throwing a few questions at him afterwards. 

Absolute clown. 

You can’t blame AI for people being shit at their job.   Copilot not the best though, so he was on an hiding to nothing.  

Oh that’s the thing, he wasn’t remotely an ‘AI’ person, just some random bloke who’s been asked by his Director to put this together. Nothing wrong with the idea, but far more qualified people in the organisation to do it. 

Chizz

PragueAddick

Been looking for something like that, thanks @Chizz

Chizz

PragueAddick said:

Been looking for something like that, thanks @Chizz

I saw it on LinkedIn this morning.  I am making no claims as to its veracity though. 

AndyG

Well it turns out A1 doesnt have a clue either when asked who will win on Sunday lol

Predicting the outcome of a sports match, particularly a playoff final, can be quite challenging due to the unpredictable nature of football. Both Charlton Athletic and Leyton Orient have their strengths and weaknesses, and the result could hinge on various factors like form, injuries, and tactics on the day. It might be beneficial to look at recent performances, head-to-head statistics, and any expert predictions leading up to the match for more insights. Ultimately, it's best to enjoy the game and see how it unfolds!

The Red Robin

PragueAddick said:

Been looking for something like that, thanks @Chizz

https://arxiv.org/pdf/2505.06120

stonemuse

AndyG said:

Well it turns out A1 doesnt have a clue either when asked who will win on Sunday lol

Predicting the outcome of a sports match, particularly a playoff final, can be quite challenging due to the unpredictable nature of football. Both Charlton Athletic and Leyton Orient have their strengths and weaknesses, and the result could hinge on various factors like form, injuries, and tactics on the day. It might be beneficial to look at recent performances, head-to-head statistics, and any expert predictions leading up to the match for more insights. Ultimately, it's best to enjoy the game and see how it unfolds!

I pressed for a prediction and got this:

Charlton Athletic 2 – 1 Leyton Orient

This scoreline reflects Charlton’s defensive strength and experience, while acknowledging Leyton Orient’s attacking capabilities. It’s anticipated to be a closely contested match, with Charlton potentially edging it in regular time.

Arthur_Trudgill

The Red Robin said:

PragueAddick said:

Been looking for something like that, thanks @Chizz

https://arxiv.org/pdf/2505.06120

Good abstract (didn't read the whole article).

Something I dislike about GenAI is that it makes it a bit too easy to blame the user for not coming up with a good prompt.

Reminds me of the ancient witch doctors when they gave you a potion to cure an illness: if it didn't work, it was because you had not used it properly, regardless of whether the potion was effective or not.

The Red Robin

Arthur_Trudgill said:

The Red Robin said:

PragueAddick said:

Been looking for something like that, thanks @Chizz

https://arxiv.org/pdf/2505.06120

Good abstract (didn't read the whole article).

Something I dislike about GenAI is that it makes it a bit too easy to blame the user for not coming up with a good prompt.

Reminds me of the ancient witch doctors when they gave you a potion to cure an illness: if it didn't work, it was because you had not used it properly, regardless of whether the potion was effective or not.

True, and I’ve definitely found it to be the case with LLMs that they get stuck in a loop of incorrect solutions if you don’t give them enough detailed information upfront. 

Chizz

https://www.instagram.com/reel/DJ9toydMnRW/?igsh=MW1wZ3NnOW5pYW9vNg==