This research introduces EnigmaToM, a novel approach that enhances large language models' ability to understand and reason about the mental states of others.

• Creates a neural knowledge base to track entity states and beliefs
• Improves efficiency in multi-hop reasoning about characters' beliefs
• Reduces reliance on LLMs for basic perspective-taking tasks
• Enables more sophisticated high-order Theory-of-Mind reasoning

Security Implications: By improving AI systems' understanding of human intent and belief states, this work could enhance security applications that need to model potential user behaviors or identify malicious intent patterns.

EnigmaToM: Improve LLMs' Theory-of-Mind Reasoning Capabilities with Neural Knowledge Base of Entity States