3. Evan Hubinger on Takeoff speeds, Risks from learned optimization & Interpretability
The Inside View
00:00
Xplaying Magic
i think most of the agen audience doesn't know about this paper anyway. This means that my audience is not literate. You can think of the behavior objective as being related to i r l, inverse reenforcement learning. In your post, you canave, gave both of the arguments and cutter arguments from christola's viewsnd, you're the best proxy, one of the best proxy of christola's view.
Transcript
Play full episode
