Wonderful article. Thanks a lot for writing in such simple and easy to follow language.
Curios, where else can monitorability or honest behaviours come from—given CoT is unreliable and Neuralese is even more dangerous?
Wonderful article. Thanks a lot for writing in such simple and easy to follow language.
Curios, where else can monitorability or honest behaviours come from—given CoT is unreliable and Neuralese is even more dangerous?