Varshul CW

• Varshul CW 8 Apr 2023 7:10 UTC

  −1 points

  0

  on: Risks from GPT-4 Byproduct of Re­cur­sively Op­ti­miz­ing AIs

  how about an having a smaller model governing safety regulations? this could act as an “aligner” on top of LLMs. say some sort of RLHF just focused on mitigating risks