04:00
2026-10-12
arxiv.org
artificial-intelligence
EvoBelief: Falsifiable World Models for Self-Improving LLM Agents
EvoBelief, a self-evolving world model that represents environment knowledge as falsifiable hypotheses, improved task performance across four benchmarks while keeping the underlying language model fro…