Neural networks are clean

Argues that interpretability can study a network's own neurons and circuits directly instead of first untangling superposition. If it holds, it changes which interpretability methods practitioners should trust and invest in.

• videoMapping the Mind of a Neural Net: Goodfire’s Eric Ho on the Future of InterpretabilitySequoia Capital
• blogCan you reverse engineer our neural network?blog.janestreet.com
• videoScaling Compute on Context — Jack Morris, EngramAI Engineer

Checking sign-in…

Loading comments…