ToM2Vec: continuous theory-of-mind vectors improve multi-agent LLM coordination
StartupYou · x · 2026-10-23
ToM2Vec teaches LLM agents to produce continuous theory-of-mind thought vectors that hold multiple hypotheses about others' beliefs, instead of only tracking their own plans. The approach improves performance across 5 multi-agent tasks and transfers to unseen partners, including frontier-model partners.
Related event: ToM2Vec: Continuous Theory-of-Mind vectors help LLM agents model teammates(10 posts)→
More from coding & agent
- Grok Bot adds 1Password integration to sign into sites without seeing your passwords — XFreeze · 2026-10-23
- How Nemesis-style systems scale: reusable rules, content costs, and why WB's patent matters — ccerrato147 · 2026-10-23
- Why Nemesis-style systems are rare: coordination costs plus WB's patent — ccerrato147 · 2026-10-23
- Spreading NPC consequences like graph traversal along relationship links — ccerrato147 · 2026-10-23
- Original Nemesis design: constrained growth and weighted randomness for limited surprise — ccerrato147 · 2026-10-23
- enRAGE mod brings Nemesis-style grudges to RDR2, with per-NPC hate/fear state tracking — ccerrato147 · 2026-10-23