A critique of AI alignment leadership’s technical grasp and threat models
One poster argues that AI alignment leaders lack practical knowledge of distributed systems and model infrastructure, and questions how they define an AI agent.
TLDR
A poster argues that AI alignment has gone “basically nowhere” after 15 years and several billion dollars. They blame leaders they say lack hands-on knowledge of computing infrastructure and criticize a risk framework they believe treats refusals, compliance and strong evaluations alike as signs of trouble.
Combined views
45
1 Source, first seen 4h ago