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

arXiv 2026 34.5 method

TLDR

Memento 3 enables frozen LLM agents to continually learn and refine explicit natural-language rulebook world models, compiled into code, for recursive self-improvement on ARC-AGI-3.

Reasoning

The paper introduces a novel model-based memory loop with rulebook revision, code compilation, and verification, plus a parallel world-model population. Weaknesses include the abstract's narrow scope and lack of visible comparison details or broader real-world validation.

Read-first score

Read-first score 34.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 40.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 25%

57.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,reproduce

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 306.

Keyword Scores

world model

world dynamics prediction

interactive world model

world simulator

model-based reinforcement learning world model

generative world model

video world model

Deep Analysis

Innovations

Introduces Memento 3, enabling frozen LLM agents to continually learn explicit world models through external memory.
Uses a natural-language rulebook as persistent semantic memory that stores revisable environment-dynamics hypotheses while leaving unknown aspects underspecified.
Compiles the rulebook into executable code for prediction and planning.
Defines a continual loop of observation, reflection, rule revision, compilation, and verification that uses prediction errors to refine the rulebook and code.
Accepts updated code only when an LLM judges it faithful to the rulebook and cell-exact replay reproduces observed transitions.
Proposes a model-based route to recursive self-improvement with a frozen underlying LLM, where verified world-model updates guide subsequent interaction and learning.
Includes a population extension that maintains multiple world models in parallel, shares interaction evidence, and uses predictions to guide exploration.

Methodology

Memento 3 equips frozen LLM agents with an external natural-language rulebook as persistent semantic memory, which is compiled into executable code for prediction and planning. The agent iterates through observation, reflection, rule revision, compilation, and verification, accepting code updates only when the LLM judges them faithful to the rulebook and cell-exact replay matches observed transitions. Evaluation uses ARC-AGI-3 public games and an Atari Pong case study, reporting Relative Human Action Efficiency and action counts.

Key Results

On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency of 100.0, and uses 44% of the human action count. In the Atari Pong case study, a learned feedback controller wins 21:0 in each of three evaluated episodes with different openings without further LLM calls.

Resources

Paper