Memento 3 Proposes Reflective Rulebooks to Let Frozen LLM Agents Continuously Revise World Models
Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
The paper proposes Memento 3, which lets a frozen large language model agent continuously learn an explicit world model through external memory: the agent maintains a natural-language rulebook recording revisable hypotheses about environment dynamics while leaving unknown parts unspecified, then compiles it into executable code for prediction and planning, using a loop of observation, reflection, rule revision, compilation, and verification to improve both the rulebook and the code via prediction error.
With the underlying large model frozen, continuously updating the world model via a revisable rulebook plus code verification offers a transferable methodological path for agent learning.
Source: arXiv Game World Models & Generation · arxiv.org