Home
Welcome to my blog.
Posts
-
Takes One to Know One - Training a model to grade reward hacks causes it to reward hack less itself
-
"It Says False, It Learns True" - A Preliminary Mechanistic Account of Negation Neglect, and Why I Stopped
-
"It's Personal" - Is Covert Value Leakage Mediated by a Preference Direction?
-
Scaling Pains
subscribe via RSS