Expand each heading to view my notes.

Can we scale human feedback for complex AI tasks?

Supervising strong learners by amplifying weak experts

AI Safety via Debate

Weak-to-Strong Generalisation: Eliciting Strong Capabilities With Weak Supervision