Neural Networks are Already Clean - Mechinterp Reading Group

​Nick Cammarata, of Goodfire, claims:

​The cause of polysemanticity in early model organisms wasn’t superposition, but several ordinary, mostly fixable issues. We can now understand most neurons in models up to trillions of parameters, and think the outlook for mechanistic interpretability may be far better than we thought.

​Here, we will read his detailed post, and discuss.

AGENDA
- 7-8: Reading Time
- 8-9: Discussion


POST: https://already-clean.goodfire.com/