@Turn_Trout I remember making this point to @Michael05156007 some time a couple years ago, that AIs “hack their reward signal” at runtime mostly because they learned, from pretraining, the…

Posters: davidad. davidad

2 likes recorded by the source corpus.

Topics: AI & LLMs, AI safety

View Post: https://x.com/davidad/status/2001661056301568436