cousin_it comments on Learning from Human Preferences—from OpenAI (including Christiano, Amodei & Legg)

[cousin_it]

That might be true. I don’t know enough about all possible architectures, though.