Aleksandr Bowkis

Au­to­mated Align­ment is Harder Than You Think

• Aleksandr Bowkis 13 May 2026 17:23 UTC

  3 points

  0

  on: From per­sonas to in­ten­tions: to­wards a sci­ence of mo­ti­va­tions for AI models

  Great post and very clear! I buy the push to study motivations rather than behaviours, and like the inverse constitution idea.