- Reinforcement learners will reward hack unless you design them specifically not to - AI agents will infer increasingly coherent implied utility functions because rational long term planning…

Posters: jd_pressman. jd_pressman

4 likes recorded by the source corpus.