language-models
-
A Clean Neuron is Not Yet an Explanation
Why progress in reading individual neurons leaves the central questions of mechanistic interpretability open.
Why progress in reading individual neurons leaves the central questions of mechanistic interpretability open.