← Back to all news
arXiv AI By Chen Wu, Josh Passenger, Yin Song

Tracing the Thoughts of a Coding Agent Playing ARC-AGI-3: Lessons for Continual Learning

Read the original on arXiv AI →

The paper investigates how a coding agent learns across a sequence of abstract reasoning tasks in the ARC-AGI-3 benchmark. The agent, operating on a frozen foundation model, writes and runs Python and shell scripts, retaining only its written artifacts as state. The authors develop a measurement protocol that traces each thought—any belief, rule, or plan committed to a file—through its lifecycle, revealing that most task-specific scripts are discarded at task boundaries and that the agent selectively forgets by rewriting knowledge into new scripts while accumulating unedited notes.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

arXiv AI
Sep 30

StateTape: Action-Conditioned Evidence Lifecycle Modeling for Long-Horizon Coding Agents

StateTape introduces a new framework for long‑horizon coding agents that rewrites the agent’s context as the code repository changes, rather than letting the context grow with every observation. It models the repository as a symbol‑level code graph, using a tape to mark symbols altered by each write and a manager model to resolve stale records. The authors provide theoretical analysis, a new benchmark called TraceBench, and empirical results showing higher resolve rates across six agents and three edit‑heavy benchmarks with minimal computational overhead.

By Ziyang Yu, Liang Zhao, Bowen Zhu, Hasibul Haque
llmsagentsbenchmarks
More like this →
arXiv AI
Oct 9

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

Memento 3 is a frozen large‑language‑model agent that learns explicit world models by maintaining a natural‑language rulebook in external memory. The agent iteratively observes, reflects, revises the rulebook, compiles it into executable code, and verifies updates through prediction error and replay, accepting only code that the LLM deems faithful to the rulebook. In experiments, a single‑model version solved all 25 public ARC‑AGI‑3 games with perfect efficiency, and a learned feedback controller achieved perfect scores in Atari Pong without further LLM calls.

By Haoyu Zhao, Zhengxu Yu, Zhiyuan He, Meng Fang, Rasul Tutunov, Haitham Bou-Ammar, Weilin Luo, Jun Wang
llmsagents
More like this →
arXiv AI
Sep 3

Belief-Calibrated Optimization: An Explicit World Model for Agentic Optimization

The paper introduces Belief-Calibrated Optimization (BCO), a method that records and updates a persistent in‑context document representing an agent’s belief about how the environment responds to edits. By continually revising this world model as new candidates are evaluated, BCO improves the performance of frozen LLM agents across five benchmarks, outperforming a control lacking the world model. An offline ablation shows that the document’s content provides reusable, accurate predictions of environmental responses, beyond mere form.

By Yuhan Chen, Zhihua Tian, Mahavir Dabas, Charith Peris, Rahul Gupta, Ming Jin, Feiyang Kang, Siyuan Zhang, Nan Wang, Ruoxi Jia
llmsagentsbenchmarks
More like this →