Whose Effect Is It? How Objectives Inflate Learned Subspace Interventions

Paper recorded by Signals 4 on 2026-11-03 in cs.CL. Abstract reproduced from arXiv; link to the original below.

Published 2026-11-03 on arXiv · recorded by Signals 4 on 2026-11-04

Category: cs.CL · 自然语言处理 · first seen 2026-11-04

Abstract

Interpretability often locates concepts by copying part of a language model's state from one input into another (activation patching). A larger output change counts as stronger evidence that this part carries the concept. Methods that learn which part to copy often change the output more than the whole state does. By that reading, learned parts isolate the concept more cleanly. Yet the whole conta

Read on arXiv →