x
How well do truth probes generalise? — AI Alignment Forum