Emergent World Beliefs: Exploring Transformers in Stochastic Games

Adam Kamel Tanish Rastogi Michael Ma University of Waterloo Independent Researcher Independent Researcher [email protected] [email protected] [email protected] Kailash Ranganathan Kevin Zhu University of California, Berkeley Algoverse AI Research [email protected] kevin@al · 2025

Read Emergent World Beliefs: Exploring Transformers in Stochastic Games by Adam Kamel Tanish Rastogi Michael Ma University of Waterloo Independent Researcher Independent Researcher [email protected] [email protected] [email protected] Kailash Ranganathan Kevin Zhu University of California, Berkeley Algoverse AI Research [email protected] kevin@al (2025) 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-2512-23722.txt.

Description

Adam Kamel, Tanish Rastogi, Michael Ma, Kailash Ranganathan, and Kevin Zhu ask what a language model secretly learns about a game it cannot fully see. In games of perfect information such as chess, models are known to build internal representations of the board. Poker is different: opponents' cards are hidden, so success requires tracking beliefs about unknown hands, not just visible state.

The team pretrained a GPT-style model on poker hand histories and probed its internal activations. Without explicit instruction, the model had learned deterministic structure, such as hand rankings, and probabilistic features, such as equity, the chance a hand ultimately wins. Nonlinear probes could decode these features, and the representations correlated with theoretical belief states. The work was presented at a NeurIPS 2025 mechanistic interpretability workshop.

Catalog record

CREATORAdam Kamel Tanish Rastogi Michael Ma University of Waterloo Independent Researcher Independent Researcher [email protected] [email protected] [email protected] Kailash Ranganathan Kevin Zhu University of California, Berkeley Algoverse AI Research [email protected] kevin@al
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-23722

Browse the full library · About the Tower of Babel