Nithin Sivakumaran, Ali Payani, Hyunji Lee, Mohit Bansal, Elias Stengel-Eskin · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.29934 · cited by 0 works (OpenAlex) · open-access PDF
La formation d'une théorie spécifique de l'agent (ToM) est la clé de la réussite de la collaboration multi-agent et nécessite un raisonnement sous incertitude pour capturer plusieurs états partenaires possibles. Cette tâche est difficile à faire dans l'espace de représentation de texte natif à grand modèle de langue (LLM) agents, qui décrivent plus facilement un seul état partenaire. Pour relever ce défi, nous introduisons ToM2Vec, qui apprend à former des théories abstraites de l'agent dans un espace latent riche à travers un budget fixe de jetons de raisonnement continu. Pendant la formation, la supervision de ces jetons provient des pensées réelles du partenaire, que les jetons ToM sont formés pour abstraire et tester. À l'heure, ToM
Machine translation (Meta M2M100); the English original follows.
Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state. To address this challenge, we introduce ToM2Vec, which learns to form abstract theories-of-mind in a rich latent space via a fixed budget of continuous reasoning tokens. During training, supervision for these tokens comes from the partner's actual thoughts, which the ToM tokens are trained to abstract and match. At test-time, ToM tokens are generated by each agent and act as conditioners for that agent's thoughts and actions. Across several environments covering collaborative, competitive, and mixed-motive settings (Overcooked-AI, multiparty negotiation, Decrypto, Hanabi, and Collab Escape/Capture), we evaluate ToM2Vec's ability to support multi-agent interactions and collaboration, finding consistent improvements over the base model and a variety of competitive trained and prompted baselines. We show that latent reasoning is critical, with performance dropping when reasoning is expressed in text. Moreover, we show that ToM2Vec training generalizes to partner model families not seen at train-time, including a proprietary model (GPT-5.4). In our analysis, we provide evidence that the latent tokens encode a ToM, rather than task-specific information: ToM2Vec's tokens improve the agent's ability to directly answer questions about partner beliefs, desires, and next actions.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Computer science, ENCODE, Representation (politics), Variety (cybernetics), Space (punctuation), Artificial intelligence
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.29934, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange