Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
cs.AI, cs.CL, cs.CV, cs.LG
Submitted: 2026-10-08
Updated: 2026-10-08
License: http://creativecommons.org/licenses/by/4.0/
The gist: Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives.
Terminology
Abstract
Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new evidence arrives. Yet limited observations can support multiple world models that explain past interactions but predict different outcomes in unseen states. We introduce Memento 3, building on the Memento series to enable frozen LLM agents to continually learn explicit world models through external memory. The agent maintains a natural-language rulebook as persistent semantic memory, recording revisable hypotheses about environment dynamics while leaving unknown aspects underspecified. It compiles this rulebook into executable code for prediction and planning. Through a continual loop of observation, reflection, rule revision, compilation, and verification, the agent uses prediction errors to refine both the rulebook and its code. Updated code is accepted only when the LLM judges it faithful to the rulebook and cell-exact replay reproduces the observed transitions. 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. A population extension maintains multiple world models in parallel, sharing interaction evidence and using their predictions to guide exploration. On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency (RHAE) of 100.0, and uses 44% of the human action count. In an Atari Pong case study, a learned feedback controller wins 21:0 in each of three evaluated episodes with different openings, without further LLM calls.
Sources
- ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence
- Belief-Calibrated Optimization: An Explicit World Model for Agentic Optimization
- OPINE-World: Programmatic World Modeling with Ontology-error-Prioritized Interactive Exploration for ARC-AGI-3
- A Review for Deep Reinforcement Learning in Atari:Benchmarks, Challenges, and Solutions
- GDI: Rethinking What Makes Reinforcement Learning Different From Supervised Learning
- NVIDIA-labs OO Agents: Native Python Object-Oriented Agents
- World Models
- Continual Harness: Online Adaptation for Self-Improving Foundation Agents
- Tycho: Active Abstraction with Programmatic World Models for ARC-AGI-3
- Code World Models for General Game Playing
- Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3?
- Executable World Models for ARC-AGI-3 in the Era of Coding Agents
- Graph-Based Exploration for ARC-AGI-3 Interactive Reasoning Tasks
- Workspace Optimization: How to Train Your Agent
- Memento 2: Learning by Stateful Reflective Memory
- Mastering Atari Games with Limited Data
- Memento: Fine-tuning LLM Agents without Fine-tuning LLMs
- Memento-Skills: Let Agents Design Agents
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems