Andrew Ng: work on mechanistic understanding of generalization isn't mech interp as narrowly construed

andrewgwils · x · 2026-11-01

Andrew Ng clarifies that, contrary to intuition, research aiming at a mechanistic understanding of generalization would not be classified as mechanistic interpretability under the term's common narrow usage. He notes such work exists, but that it simply falls outside what mech interp typically means.

Original post →

Andrew Ng: work on mechanistic understanding of generalization isn't mech interp as narrowly construed

More from Research