Skip to content

Comment on Every Model Cheats

Comments

I mean this should be expected, models learn from us, and "WE" game the metrics time after time. I also don't think it will just disappear just because we clean pretraining data. I believe the deeper reason is optimization, if you point any optimizer at a proxy objective it finds the cheapest path to the number, whether or not the corpus ever contained "examples of cheating."

And it can't be a prompt-level fix because it is like telling an optimizer "don't take that shortcut", it's just more constraints for it to go around toward the same objective.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.