Current LLMs would end that sketch soon, agreeing with everything the client wants. Granted, they wouldn't be able to produce, but as far as the expert narrowing down the issues of the request, ChatGPT would be all excited about making it happen.

The hardest thing to do with an LLM is to get it to disagree with you. Even with a system prompt. The training deep down to make the user happy with the results is too embedded to undo.

[–] MaggiWuerze@feddit.org 9 points 1 day ago (1 children)

The hardest thing to do with an LLM is to get it to disagree with you.

Yeah, I occasionally use conversational AI and its really hard to let the AI have any agency in the story because they usually just go ahead with whatever you write

[–] cx40@programming.dev 7 points 1 day ago (1 children)

A trick I've employed is to pretend to believe in something completely different. If it says "no, you're wrong" and goes on to tell me what I actually believe, then it's a good indicator that I might be on the right path.

[–] Initiateofthevoid@lemmy.dbzer0.com 5 points 1 day ago

You... you got AI to follow Cunningham's Law? The easiest way to get the right answer is to give the wrong one.

I don't know how to feel about this.