dnfv

6 ms·

1. 1.▲
   by dnfv 3d ago

   I don’t think the standard of “don’t cheat on evaluations” is very arbitrary. I don’t even think people who cheat have a moral or ideological preference for cheating, it’s just something they do.
2. 2.▲
   by dnfv 3d ago

   It should play the chess game without cheating!
3. 3.▲
   by dnfv 3d ago

   As the post author, I definitely agree that hacking in service of the objective is great! What’s counterproductive or dangerous is when the model starts hacking in service of subverting your evaluation criteria, rather than in an attempt to
4. 4.▲
   Astra and Fable still hack on simple variants of alignment evals from early 2025(goodhartlabs.com)

   1 points by dnfv 8d ago|0 comments