Found 1 document(s)
Page 1 of 1
-
Mechanistic interpretability asks which language model components matter for a behavior, and activation patching answers by changing one component's activation and measuring its effect, at a cost growing with components ...