1. put model in an easily reward hackable environment 2. let it reward hack for 600 steps 3. make a steering vector of ckpt-600 <> original 4. steer original model very heavily on this vector…
Posters: voooooogel. voooooogel
5 likes recorded by the source corpus.
Topics: AI safety
View Post: https://x.com/voooooogel/status/1970578046643294353