EvoBelief checks and changes the environment rules of an LLM agent
In DiscoveryWorld Reactor Lab, it completed 83.3% of Normal-tier instances and 58.3% of Challenge-tier instances, and each baseline completed 0%.
Claimed, not confirmed
The researchers made EvoBelief, a self-evolving world model for LLM agents. The model keeps knowledge of the environment as hypotheses that a test can show to be incorrect. A hypothesis becomes a rule only after subsequent tests show that it is correct. A counterexample limits or removes the rule. The LLM stays frozen. The model parameters do not change.
Sources
Posted
