I'm an AI safety researcher, focused on weight-based interpretability.
Discuss some idea in depth.
A quick chat.