paper / owainevans_uk / May 12 / failedSchool of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMs
paper / owainevans_uk / May 12 / failedWeird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
paper / owainevans_uk / May 12 / failedThe Consciousness Cluster: Emergent preferences of Models that Claim to be Conscious