Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks

Key points

• Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives.1
• Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states.1
• The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified.1
• We investigate this process as a model-based route to recursive self-improvement (RSI): the agent autonomously explores the environment, revises its world model, and uses verified updates to guide subsequent interaction and learning, while the underlying LLM remains fixed.1

Sources (2)

• [1]Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
  Hugging Face Daily Papers · Oct 8, 12:00 AM

  We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory.

  Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives.

• [2]Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 12:06 PM · same content

Extractive summary: sentences quoted from the sources.