TurnTrout comments on Modulating sycophancy in an RLHF model via activation steering

[TurnTrout]

Actually, Towards Understanding Sycophancy in Language Models presents data supporting the claim that RL training can intensify sycophancy. EG from figure 6