2 papers
cs.CL2025
Decomposing Attention To Find Context-Sensitive Neurons
Alex Gibson
We study transformer language models, analyzing attention heads whose attention patterns are spread out, and whose attention scores depend weakly on content. We argue that the soft…
cs.LG2024
Compact Proofs of Model Performance via Mechanistic Interpretability
Jason Gross, Rajashree Agrawal, Thomas Kwa +5
We propose using mechanistic interpretability -- techniques for reverse engineering model weights into human-interpretable algorithms -- to derive and compactly prove formal guaran…