#if you suggest improving the prompt ur getting blocked. ive refined the prompt provided examples and instructed it multiple times. i tried.

[lensdeer]

I've been begrudgingly using ChatGPT for work for about a week now. My boss (to whom I have repeatedly explained how LLMs work and why they don't fit our usecase too well, but I digress) is getting inceasingly annoyed I don't use it and write my own shit instead. Figured I could get rid of some heat by using it for ONE project, like, I can say I "tried" it and can go back to actually doing things properly after this is done.

So I decided to use it as "spicy regex": I gave it multiple examples of how to convert structured plaintext data from a public website into a JSON; nothing I couldn't do with a sufficiently long regex string. This is legitimately the perfect use case for a LLM! Yet it's fucking failing at least once for EVERY SINGLE INPUT!

Repeatedly forgetting instructions, hallucinating random bullshit, giving answers that have nothing to do with my request, giving me incomplete outputs and ignoring direct instructions to complete them just to character-for-character repeat the same mistake it just did...

The moral panic and misinformed environmentalist fearmongering have obviously been useless at taking this thing off the pedestal tech oligarchs paid to get it put on, but, like,,, It's honestly MISERABLE to use??? You don't need a moral panic, like, just make someone use this thing for something they're good at and watch them slowly realize how fucking awful it is. How the fuck do AI bros cope.

At this point I've wasted enough time repeating instructions on how to do shit correctly that it would've been MUCH faster to write a script to grab all the information and format it without any user input. But nope. Gotta finish the project with this useless fucking toy or eventually get fired for "not catching up to new technologies". Lovely.