Localizing Anchoring Pathways in Language Models

📰 ArXiv cs.AI

arXiv:2606.12818v1 Announce Type: cross Abstract: Irrelevant numbers in a prompt can shift language model judgments, producing anchoring effects in numerical reasoning. We study where this anchor-sensitive signal is carried inside language models using a controlled multiple-choice setup with shared answer options. We define a logit-difference metric comparing the correct answer option with the answer option corresponding to the anchor, and validate that it tracks behavioral anchoring. Using attr

Published 12 Jun 2026

Full Article

Title: Localizing Anchoring Pathways in Language Models

Abstract:
arXiv:2606.12818v1 Announce Type: cross Abstract: Irrelevant numbers in a prompt can shift language model judgments, producing anchoring effects in numerical reasoning. We study where this anchor-sensitive signal is carried inside language models using a controlled multiple-choice setup with shared answer options. We define a logit-difference metric comparing the correct answer option with the answer option corresponding to the anchor, and validate that it tracks behavioral anchoring. Using attr
Read full paper → ← Back to Reads