Theory of mind (ToM) is the capacity to reason about others' mental states, and it underlies effective communication and coordination. Higher-order ToM extends this to nested perspectives, such as reasoning about what another person believes about your beliefs, which matters in strategic interactions.
A new arXiv paper, "Beyond Belief Nesting: Rethinking Higher-Order Theory of Mind in Large Language Models," takes up this topic in the context of LLMs. The abstract frames higher-order ToM as a key capability for strategic interaction, but suggests that current approaches—centered on belief nesting—may need to be reconsidered.
The full details of the proposed rethink are not available in the abstract excerpt. The paper is posted on arXiv under identifier 2610.17777, with an announcement type of "new."