Attention: Please take a moment to consider our terms and conditions before posting.
Chat GPT and other AI guff
Comments
-
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.cantersaddick said: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.0 -
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.
2 -
Yeah, it's mega...Arthur_Trudgill said: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.
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'3 -
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.6 -
I don't love GenAI, but I find it useful for some things.Leroy Ambrose said:
Yeah, it's mega...Arthur_Trudgill said: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.
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'
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.2 -
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.0 -
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.3 -
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🤣.1
-
I use CGPT to make marketing flyers, also has been very useful in breaking down some extremely elongated excel formulas for my KPI monitors0
-
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.2
-
Sponsored links:
-
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 up3
-
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.DamoNorthStand said: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
2 -
This certainly resonates with a lot of what I see where I work…
https://apple.news/AcoDGlu3zREGlXDlAYQxo5A
0 -
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).SuedeAdidas said:This certainly resonates with a lot of what I see where I work…
https://apple.news/AcoDGlu3zREGlXDlAYQxo5A
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.4
