366

submitted 3 days ago by FoxtrotDeltaTango@sh.itjust.works to c/programmer_humor@programming.dev

87 comments fedilink hide all child comments

you are viewing a single comment's thread
view the rest of the comments

[-] definitemaybe@lemmy.ca 75 points 3 days ago* (last edited 3 days ago)

When you're vibe coding, you need to narrate long prompts, including things like "And remember not to hallucinate or make any errors." All that is a lot faster to say than type.

Clearly, you're not vibe coding, and your job will be (checks notes) automated away by someone with a #grindset.

/s, in case that wasn't obvious.

[-] DisasterTransport@startrek.website 25 points 3 days ago* (last edited 3 days ago)

If youre not using a custom instruction that brute forces different ways to try to get the other model not to fuck uo youre doing it wrong. If that doesn't work, add another llm into the pipe

Edit what is a token and why is my boss sending me angry emails

[-] Croquette@sh.itjust.works 17 points 3 days ago

I was talking to an acquaintance and she was taking AI course.

She was taught to "validate" LLM infos by asking the same question in a different model.

This is an actual university course.

I was baffled. There is no guarantee that both models aren't wrong. But speaking to people that use LLMs, this is a common strategy.

[-] Goodeye8@piefed.social 19 points 3 days ago

Actually the tooling is such that you have that written in some kind of a rules file that will then more or less append the rules to the prompt so you wouldn't have to repeat it over and over.

At least I think that's how it works, I was only half listening to the mandatory AI training.