Attention: Please take a moment to consider our terms and conditions before posting.

Chat GPT and other AI guff

11213141618

Comments

  • BalladMan
    BalladMan Posts: 1,460
    Agree about Claude. GPT grabbed some market share by being first but is by a long way the worst of the AI tools.

    Though I generally haven't found much use for any of them. Massive flaws every time I have tried to use them for anything useful. And the studies into cognitive decline in people who rely on AI is scary - basically replicating dementia in those who over rely on it.
    Not sure I ageee on OpenAI more generally.  ?codex (tool anyone can use to build anything technical) is heads and shoulders above its rivals.  ChatGPT makes OpenAI no money. They need the enterprise market to start turning around massive losses which is where Codex comes in. 
  • Arthur_Trudgill
    Arthur_Trudgill Posts: 597
    We use Claude Code from Anthropic, and it works well as a coding assistant.

    It's good for generating experimental software that I would not have written because it was just one of many possibilities, not all of which I would have had the time to try out.

    But I've noticed you have to know quite a bit about software to be able describe what you want it to write in the first place, and check it once finished.

    I imagine it is the same for most uses: it's an assistant to someone who already knows quite a bit about the subject.


  • Leroy Ambrose
    Leroy Ambrose Posts: 14,809
    edited June 8
    We use Claude Code from Anthropic, and it works well as a coding assistant.

    It's good for generating experimental software that I would not have written because it was just one of many possibilities, not all of which I would have had the time to try out.

    But I've noticed you have to know quite a bit about software to be able describe what you want it to write in the first place, and check it once finished.

    I imagine it is the same for most uses: it's an assistant to someone who already knows quite a bit about the subject.


    Yeah, it's mega...  :*



    Literally just now, while refreshing a dashboard I built from a bunch of spreadsheets - I got this after a previous failed attempt to debug a simple js function. Silly me though, I forgot to give it the correct context in my prompt: "Claude, fix these errors and DON'T FUCKING GUESS'
  • mascot88
    mascot88 Posts: 10,090
    edited June 8
    Claude is a professional life improver for me - I'm in digital marketing so - AEO, SEO, Google Ads, META , Conversion Rate Optimisation - that sort of stuff. 

    I can connect Claude to various tools and get it to build strategies in minutes that would have taken me hours perviously - I can then stress test them I can trim them, make adjustments and have Claude do it all. 

    I can then have Claude package them in a deck ready to present - in my own brand colours. 

    It is literally optimising my performance and saving me large chunks of time. 

    I love it. 
  • Arthur_Trudgill
    Arthur_Trudgill Posts: 597
    We use Claude Code from Anthropic, and it works well as a coding assistant.

    It's good for generating experimental software that I would not have written because it was just one of many possibilities, not all of which I would have had the time to try out.

    But I've noticed you have to know quite a bit about software to be able describe what you want it to write in the first place, and check it once finished.

    I imagine it is the same for most uses: it's an assistant to someone who already knows quite a bit about the subject.


    Yeah, it's mega...  :*



    Literally just now, while refreshing a dashboard I built from a bunch of spreadsheets - I got this after a previous failed attempt to debug a simple js function. Silly me though, I forgot to give it the correct context in my prompt: "Claude, fix these errors and DON'T FUCKING GUESS'
    I don't love GenAI, but I find it useful for some things.

    It takes skill to use it effectively, like most tools, so the Intelligence bit is over hyped; I would have preferred a term like 'smart', or some such.

    For coding, the annoying thing is that you have to review carefully what is has coded, and yet code reviewing is one of the most tedious aspects of IT.
  • Swisdom
    Swisdom Posts: 15,168
    i use claude for more digging deeper into stuff and chat gpt for more creative stuff.  They are both very impressive but neither are perfect.
    If I ask Claude to do anything creative it's sadly lacking
    If I ask Chat gpt to do anything with a consistent style it will always do it perfectly once and then i can change one tiny thing and it completely fucks it up.  
    I find myself swearing at them almost daily but the stuff it does well it does really bloody well.  
  • Leroy Ambrose
    Leroy Ambrose Posts: 14,809
    Got my hands on Mythos a while ago as part of an insider preview. It's good, but unless you've got a bottomless pit of money, the cybersecurity horror stories have been vastly over exaggerated. The nonsense I saw about it finding hundreds of Critical exploits that other tools missed was, predictably, lacking context. Most of those issues were, somewhat ironically, from AI generated code that hadn't been properly reviewed... 😏

    When let loose on some of our own applications, it performed marginally better than existing agentic tools, but we'd have burned through probably five times the tokens that we would have using Opus (where it shines is in the ability to generate attack paths and chain vulnerabilities to make their impact worse - and that requires a LOT of exploratory probing of the code, along with context on things like the environment it's deployed to, or the IAC/Container orchestration (basically, the equivalent of servers and networks from the old days) in place. 

    In short - once again, Anthropic prove that their marketing is second to none. Even having Trump pull it as part of the ludicrously thinly-veiled extortion racket the US has now become has helped their marketing campaign. 
  • PragueAddick
    PragueAddick Posts: 22,615
    This is such an educational thread, guys. Thank you. I certainly fell for the Anthropic hype around Mythos, so it's a rather delicious reality check to learn that this apparently dangerously capable new release, which scared Trump into trying to restrict access to it, has in fact been in the hands of a Charlton Lifer for several weeks🤣. 
  • DA9
    DA9 Posts: 11,193
    I use CGPT to make marketing flyers, also has been very useful in breaking down some extremely elongated excel formulas for my KPI monitors
  • swords_alive
    swords_alive Posts: 4,962
    edited June 27
    I used it for advice to mend a relationship difficulty and took it's advice not to do anything, other than 'work on myself'. It seemed comprehensive and made good sense at the time but now wondering if i should not have listened to that, or at least got it validated by someone!  The other party will have probably moved on and best leave sleeping dogs lie, but a niggling doubt recurs now and then.

  • Sponsored links:



  • DamoNorthStand
    DamoNorthStand Posts: 13,012
    Be careful about ChatGPT (and any AI module) hallucinating. It is a real issue at the moment. We asked it to pull TikTok handles for 10,500 sports and entertainment properties for our platform to scrape from..... we have had to re-do c30% of them manually as it was making all kinds of shit up
  • PragueAddick
    PragueAddick Posts: 22,615
    Be careful about ChatGPT (and any AI module) hallucinating. It is a real issue at the moment. We asked it to pull TikTok handles for 10,500 sports and entertainment properties for our platform to scrape from..... we have had to re-do c30% of them manually as it was making all kinds of shit up
    The single mundane topic which has delivered me the most nonsense is football. So many factual inaccuracies which I could spot but someone less invested in footie would have missed. The worst example was “incomings and outgoing players at Sunderland for the last 3 seasons.” How hard is that? Admittedly I was using Perplexity which I have since abandoned, but still. i kept pointing out mistakes, and it kept coming back with new ones. I gave up.  I suppose it may be a case of “garbage in” , so much footie stuff in the web is full of loose or made-up crap, as we know.

  • SuedeAdidas
    SuedeAdidas Posts: 8,137
    This certainly resonates with a lot of what I see where I work…

    https://apple.news/AcoDGlu3zREGlXDlAYQxo5A
  • Leroy Ambrose
    Leroy Ambrose Posts: 14,809
    edited June 27
    This certainly resonates with a lot of what I see where I work…

    https://apple.news/AcoDGlu3zREGlXDlAYQxo5A
    We're a software company, and the mistakes I see supposedly skilled developers making would make your toes curl, let alone the staff who've been let loose on it with no development experience at all. Luckily we out good guardrails in place early on - but still now we see shit like repos (source code) being cloned locally and unapproved models running on them, data repurposed and shared with unprivileged people and attempts to use Deepseek (Deepseek, ffs - like there haven't been enough warnings already about Chinese AI).

    The tide is starting to turn in my industry though. Companies are realising that having an entire swathe of junior devs with no skills other than prompt engineering whole mid-level engineers leave because they're pissed off at becoming code reviewers instead of engineers is a recipe for disaster. The AI companies are getting shitty with pricing now too - so a lot of the 'Hey look! I made a poorly working, buggy dashboard which already exists elsewhere in the business for the same data!' shit is starting to get shut off because spending 4000 quid on tokens for some sales wonk to brag to the company isn't particularly good value.

    I expect it'll be the same in the creative industries as people start to realise that creating marketing materials that look exactiy the same as every other fucker's isn't great in an industry where the major differentiator is... How different your marketing is 😏

    Whichever of OpenAI or Anthropic wins the IPO jerkoff battle will be interesting. Claude is miles ahead of Codex, but Altman is a filthy shithouse - so fully expect them to file first. They're both racing against time before the arsehole falls out of the market - if there's a way they can make shareholders carry the bag they will. 
  • Callumcafc
    Callumcafc Posts: 66,951
    edited July 21
    Claude Fable 5 supposedly disproves the Jacobian Conjecture, posed in 1939.

    https://mashable.com/tech/anthropic-fable-5-disproves-jacobian-conjecture
  • Callumcafc
    Callumcafc Posts: 66,951
    edited July 21
    https://en.wikipedia.org/wiki/Jacobian_conjecture

    On July 19, 2026, Anthropic employee and mathematician Levent Alpöge presented an explicit counterexample to the conjecture in three-dimensional space, stating that it was discovered using the large language model Claude Fable 5.[6][7][8] 

  • Leuth
    Leuth Posts: 24,172
    Wow, this really makes up for all the other stuff 
  • I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
  • Leuth
    Leuth Posts: 24,172
    This is the worst timeline
  • Leroy Ambrose
    Leroy Ambrose Posts: 14,809
    I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
    The tragic thing is plenty of people would have just done it and watched their grass bleached out of existence. We're sleepwalking into Wall-E levels of compliant ennui

  • Sponsored links:



  • Chizz
    Chizz Posts: 28,774
    I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
    The tragic thing is plenty of people would have just done it and watched their grass bleached out of existence. We're sleepwalking into Wall-E levels of compliant ennui
    Maybe it's a good thing.  Maybe we want not to be surrounded by unquestioning people who would act on what is so obviously wrong. 

    If ChatGPT is your first and only port of call for advice on gardening (or health issues, or tax, or relationships, or cocktail recipes, or decorating, or exercise, or investments) and you trust it without question, perhaps you don't deserve optimal outcomes.  

    It's a terrific tool, for some jobs.  If you use it for the wrong job, or in the wrong way, then you really ought to blame yourself, rather than "AI".  It's like using a spreadsheet to edit photos.  Or using a calculator as a search engine.  Or using GPS to write a novel. 

    Ricky Gervais said it well.  If we'd taken the "do not consume" warning signs off bottles of bleach, we wouldn't have voted to leave the EU. 
  • SuedeAdidas
    SuedeAdidas Posts: 8,137
    Claude Fable 5 supposedly disproves the Jacobian Conjecture, posed in 1939.

    https://mashable.com/tech/anthropic-fable-5-disproves-jacobian-conjecture
    Noooooooooooooo. 
    If we can’t believe in the Jacobean Conjecture what can we believe in?!?
  • SuedeAdidas
    SuedeAdidas Posts: 8,137
    Chizz said:
    I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
    The tragic thing is plenty of people would have just done it and watched their grass bleached out of existence. We're sleepwalking into Wall-E levels of compliant ennui
    Maybe it's a good thing.  Maybe we want not to be surrounded by unquestioning people who would act on what is so obviously wrong. 

    Like the people who drove their cars into rivers or into the sea because their satnav told them to…
  • Chizz
    Chizz Posts: 28,774
    Chizz said:
    I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
    The tragic thing is plenty of people would have just done it and watched their grass bleached out of existence. We're sleepwalking into Wall-E levels of compliant ennui
    Maybe it's a good thing.  Maybe we want not to be surrounded by unquestioning people who would act on what is so obviously wrong. 

    Like the people who drove their cars into rivers or into the sea because their satnav told them to…
    Exactly
  • Diebythesword
    Diebythesword Posts: 951
    Chizz said:
    I've battled a fungal infection in my grass for years and turned to AI. 

    Amongst other things it suggested using 3% hydrogen peroxide diluted 1 to 4.  I noticed that the 12% peroxide would work out cheaper and asked if it would be ok if diluted correctly.  It came back and said it was fine and suggested a 1 to 2 dilution.  I queried this saying it's a lot stronger than the first suggestion.  AI then said something like, "That's a great spot Raith' and gave me another made up number instead.

    Worse case scenario it could have killed my grass, I'll not consult any time soon about it my medication doseage  :s
    The tragic thing is plenty of people would have just done it and watched their grass bleached out of existence. We're sleepwalking into Wall-E levels of compliant ennui
    Maybe it's a good thing.  Maybe we want not to be surrounded by unquestioning people who would act on what is so obviously wrong. 

    Like the people who drove their cars into rivers or into the sea because their satnav told them to…
    Exactly. Treat AI like it’s a well meaning but naive junior employee on the subject you’re getting it to work on and it’s incredibly useful. Building entire software with it and stuff handling sensitive info then you deserve what happens tbh. 
  • Huskaris
    Huskaris Posts: 9,982
    I have a colleague who is on our leadership team with me, and she uses AI for literally everything.  

    There's been relatively simple projects that she has turned into 30 page word documents with 20 slide decks associated with them that basically tell you next to nothing, and grossly overcomplicate what the issue is because they don't have the full context of the person inside the business. 

    It's insanely infuriating to have to engage with it, but because it's relatively well polished, not many people are seeing through it yet. 

    On the flip side, I have used AI personally (Gemini, good free thresholds) to do useful things like calculate what my retirement pot will be in a way a pension calculator couldn't because of what I'm planning on doing etc, it was really useful for that. 

    I've also build an agent on copilot that will take legal documents with red lines from clients and create a clause by clause change summary table which has massively reduced time taken to analyse contracts, so that was useful too.

    I think there's absolutely a place for it, and I also think it's only going to get more powerful, much more powerful. 
  • cantersaddick
    cantersaddick Posts: 18,482
    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.
  • Huskaris
    Huskaris Posts: 9,982
    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.
    I vibe coded a cashflow forecasting app which old then be used to order stock (completely for fun, not even aiming to use it). What I realised is, it can create you a shell, and a user interface, but what is in the box is just shit in most cases. 

    We are building an AI solution at work and investing $4m in it pre revenue, the external coders say that they use AI to help evaluate the code they write, rather than to write the code...
  • cantersaddick
    cantersaddick Posts: 18,482
    edited July 22
    Huskaris 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.
    I vibe coded a cashflow forecasting app which old then be used to order stock (completely for fun, not even aiming to use it). What I realised is, it can create you a shell, and a user interface, but what is in the box is just shit in most cases. 

    We are building an AI solution at work and investing $4m in it pre revenue, the external coders say that they use AI to help evaluate the code they write, rather than to write the code...
    Yeah I use AI to help me code, though I need to code much less these days due to having quite a bit team but when I do either I code and it checks or it codes and I check - but I have a background in programming and data engineering. It can definitely speed things up but you need the starting skills and knowledge of how it works. It does make things up and also when it writes stuff its incredibly inefficient. Its people with no prior knowledge or skills using it to try and create products to try and make money from that annoy me. They are inevitably horse shit.
  • ShootersHillGuru
    ShootersHillGuru Posts: 51,455
    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.