What's annoying me today

We're actually already at a point where the AI companies are so deeply leveraged and the shareholders impatient for returns that the cost of tokens is making businesses start to question whether free rein for employees to use it at will is a good idea. Saving employee costs to just replace it with AI consumable fees is a decision lots won't make.

I'm trying not to be rude, but that is totally at odds with the majority of the people writing code nowadays. Have you used the latest generation of Gemini or Claude? Their coding skills are ridiculous. I open blank python scripts now and just have natural language conversations about what we need to do and it does top notch work. You might want to QA it, but honestly you can just ask it to then write it's own QA and it'll write and run all sorts of tests then fix them itself. People will still be coding in 5-10 years time, but the method is going to be totally different.

100% agree with this. We've now got a $50 a month limit for every employee with co-pilot, but you can request $150 a month if you can write a justification. Or a project specific budget if you think it warrants it. This has all come in the last two months.

We have our own internal AI built on top of ChatGPT I believe, which can be used unlimited, but it always feels a bit behind and not as good. They've just released an internal agentic tool too which I used to help be make a project pitch earlier. It created graphs, PowerPoint, Word the lot.

We have our own as well because as a federally regulated business we have a lot of internal data we aren't legally allowed to make available to a publicly accessible engine, but we're a multi billion dollar company that can afford to run our own AI instance. A lot of companies can't.

Yeah tbh same about the data. So much client data that we cant be uploading to the big providers. But as you say, I bet a ton of people are doing.

Be rude, you fucking twat. I quite clearly said "I", ffs. I didn't try to explain my job to you.

"writing code is easier than working out what I want it to do"

What do you mean then? You don't think about what you want to do before starting to write code?

I try to minimise false positives while catching the most fraud. The code is largely copy and paste, but complex. The data and data structure are non-standard with piles of exceptions. And it has to be correct. I have to work around technology limitations, staff resource, antagonistic DBAs and an absence of budget.
Plus, you could just believe me. Or else I'll presume you're just another brainwashed AI pumper who wants to justify their own laziness. And no, I absolutely do not believe that about every AI user. I can see its use for many people. Especially when coding.

You have to admit that what you're saying sounds like what most people would say if they felt their role was threatened by AI (which is pretty much everyone)!

Everyone's roles are under threat from AI, whether because their jobs are kind of unnecessary, or because management think they should be replaced. Or anywhere in between. A lot of stuff I do should be automated anyway, whether by AI or competent IT.
Either way, specialist knowledge shouldn't really be underestimated.

I agree. But it will be. And noone will know what they are missing because they will just trust the machine.

Everyone has great, noble human in the loop policies currently, but that will soon go when more people savings are needed to pay for the tokens.

Without wanting to cause an argument - if you want a range of AI options for different things, there's quite a collection here. huggingface.co/

Those are the guys who were "hacked" by OpenAI, right?
I don't need AI, fortunately, and I don't want it either. If it's going to exist, then we need regulation on how it is run, at the very least.

Shitloads.

Most people don’t write particularly well. That’s not a criticism… we’re not really taught to, more often we’re taught how to write badly (to sound ‘academic’). So along comes AI and everyone says ‘wahoo, you can use this to help you write’, and people do. We’ll often be reminded to check what the AI does (“AI first, human last”) but I don’t think all that many people do that, and of those that do… well a lot of them are using it because they can’t write well and that probably means they’re not super equipped to assess what the AI shat out.

LLM’s are shitty writers out of the box, but better prompting and reviewing can get ok results… but I’m not sure what the point is because a good writer can get better results faster by just writing, and a bad writer lacks the skills to get good writing out of the LLM. The net result is that nothing changes for the good writers, the bad writers think they’re delivering good writing but they’re not, everyone is having to read generic padded-out slop everywhere, and it’s cheaper to replace a kidney than a stick of RAM.

We’re a provider to businesses who deal with consumers and collect/hold staggering amounts of sensitive personal data (much of which ends up living on our servers). We can tell them what they should and shouldn’t be doing with it all but a whole bunch of them are definitely feeding stuff into ChatGPT global that they should not be doing because the regulators are nowhere near getting a handle on this.

I use Co Pilot to feed in loads of information to give me brief reports for meetings. I can only use the licensed version on work systems as apparently it’s more secure……

However as time saving as it is it does need to be used with caution, I started to read something out last week that I knew wasn’t correct so had to quickly change track.