Introducing the AI Alignment Forum (FAQ)

Corrigibility as Constrained Optimisation

[Question]Best reasons for pessimism about impact of impact measures?

Alignment Newsletter One Year Retrospective

Value Learning is only Asymptotically Safe

Reinforcement learning with imperceptible rewards

Alignment Newsletter #52

Defeating Goodhart and the "closest unblocked strategy" problem

Alignment Newsletter #51

Learning "known" information when the information is not actually known