logoalt Hacker News

colechristensentoday at 12:54 AM1 replyview on HN

No, they just need to be trained to have adversarial self review "thinking" processes.

You ask an LLM "What's wrong with your answer?" and you get pretty good results.


Replies

binary0010today at 12:59 AM

Or you get the original output result was perfect and the adversarial "rethinking" switches to an incorrect result.

show 1 reply