Causal Localization of Evidence Effects in Transformers
Can controlled symbolic inputs reveal attention heads that influence a language model's decisions from natural language evidence? We introduce controlled evidence aggregation (CEA), which varies the counts and order of two symbols in otherwise fixed prompts and measures changes in a frozen…
By Jipeng Li, Pramod Khargonekar, Yanning ShenarXiv - cs.CL (NLP)
Read the full story
Originally published by arXiv - cs.CL (NLP) on Oct 28, 2026.
Comments
Sign in to join the discussion.