Google presents AtP

An efficient and scalable method for localizing LLM behaviour to components.

Activation Patching is a method of directly computing causal attributions of behavior to model components.

---

Join the discussion on this paper page.