Indications of Belief-Guided Agency and Meta-Cognitive Monitoring in Large Language Models

Noam Steinmetz Yalon, Ariel Goldstein, Liad Mudrik, Mor Geva · 2026

Read Indications of Belief-Guided Agency and Meta-Cognitive Monitoring in Large Language Models by Noam Steinmetz Yalon, Ariel Goldstein, Liad Mudrik, Mor Geva (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02467.txt.

Description

Noam Steinmetz Yalon and 3 co-authors test one proposed indicator of consciousness in large language models, in a 2026 arXiv paper. The indicator, drawn from a neuroscience-based checklist, asks whether agency is guided by belief formation that updates through meta-cognitive monitoring, meaning the system watches its own states. The authors treat beliefs as representations in the model's latent space and define a metric for their dominance during generation.

Across models and tasks, three patterns appeared. External manipulations systematically shifted internal belief formation. Belief formation causally drove which actions the model selected. Models could also monitor and report their own belief states. The authors read these results as empirical support for belief-guided agency under that indicator. They stress the methodological contribution: tools for probing agency and meta-cognition in these systems.

Catalog record

CREATORNoam Steinmetz Yalon, Ariel Goldstein, Liad Mudrik, Mor Geva
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02467

Browse the full library · About the Tower of Babel