Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelswith Carson Denison, Monte MacDiarmid, Fazl Barez, David Duvenaud +92 versionsW_8zqhdax4v2 · current