Hacker News new | ask | show | jobs
by binary0010 36 days ago
Or you get the original output result was perfect and the adversarial "rethinking" switches to an incorrect result.
1 comments

this seems to happen far more than i would like