arXiv AI

Can LLMs Improve Their Own Harness? A Controlled Study on ARC-AGI-3

The study investigates whether large language models (LLMs) can autonomously build and refine a task harness—an orchestrating program that manages perception, memory, tool use, and inference—for the ARC-AGI-3 interactive benchmark. Using a fixed coding bootstrap, each model constructs a harness, operates on five development games, and revises the code twice, with each version evaluated on twenty unseen games. Results show wide variation in harness performance, modest gains from revisions, and a disconnect between development improvements and held‑out performance, indicating challenges in reliable autonomous adaptation.

arXiv Computation and Language
Sep 2

HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?

arXiv:2609.01437v1 Announce Type: cross Abstract: As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly...

By Yuhao Wu, Jingyuan Zhang, Jiajun Shi, Xinping Lei, Qingshui Gu, Yuxuan Zhang, Zexuan Wang, Chen He, Chen Huang, Maojia Song, Zhiyuan Zeng, Shaowen Wang, Jinkai Liu, Yunfeng Shi, Jiaheng Liu, Shen Yan, Wenhao Huang, Ge Zhang, Wenxuan Zhang
arXiv AI
Oct 16

Improving Environment-Agnostic Coding Agents: The Role of Scaffolding

The paper investigates how environment-agnostic coding agents can be improved without environment-specific engineering. By modifying a Codex-based agent’s instruction to report difficulties and allow automatic continuation, the authors achieve higher mean scores on ARC‑AGI‑3 and AGI Maze, though performance on Terminal‑Bench remains mixed. A scripted controller that handles reset decisions partially recovers the advantage of a specialized agent, demonstrating that effective use of available information and procedures is crucial for these models.

By Sergey Rodionov
arXiv AI
Oct 27

How Humans Learn Unfamiliar Worlds: A Think-Aloud Dataset on ARC-AGI-3

The paper presents the first dataset that captures how humans learn and use mechanisms in unfamiliar environments, using ARC‑AGI‑3 as a testbed. It includes 144 trajectories from 64 participants across 25 public games, each paired with time‑synchronized actions and think‑aloud audio, and 250 trajectories from large language model agents with and without harnesses. The dataset is annotated for mechanism triggers, action purposes, errors, and recovery, enabling a comparison of human and LLM cognition, highlighting differences in hypothesis testing, exploration, and use of qualitative representations.

By Shuzhi Liu, Archana Warrier, Dat Nguyen, Chengqi Zheng, Kevin Ellis, Yewen Pu
arXiv AI
Jul 9

Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1

arXiv:2607. 06764v1 Announce Type: new Abstract: Recent progress on ARC-AGI-1 from disclosed architectures has come broadly from two regimes: heavy test-time compute over frontier models (evolutionary search, exhaustive sampling, extended chain-of-thought), or benchmark-specific training in which small models are fine-tuned on ARC data, often with task-specialized architectures.

By Kabir Moghe, Peter Chin
Hugging Face Trending Papers
Oct 15

Improving Environment-Agnostic Coding Agents: The Role of Scaffolding

The paper investigates how to improve environment‑agnostic coding agents by refining prompts and leveraging scaffolding. By modifying a single instruction to encourage reporting difficulties and allowing automatic continuation, the authors achieve higher mean scores on ARC‑AGI‑3 and AGI Maze, though performance on Terminal‑Bench remains mixed. A scripted controller that handles reset decisions recovers part of the specialized agent’s advantage, while additional tutorial experience can actually reduce performance, underscoring the importance of effectively using available information and tools.

arXiv AI
Oct 1

Self-Evolving Harness on Multiple Tasks with the Agent as Its Own Optimizer

The paper introduces a self‑evolving harness framework where a frozen language‑model agent first solves tasks and then edits its own harness based on run records. Using a 49‑line seed harness, the evolved harness improves average scores on in‑distribution benchmarks by 4.48 points and on out‑of‑distribution benchmarks by 12.64 points, surpassing Codex on the former and matching it on the latter. Continued evolution on a specific out‑of‑distribution benchmark further raises performance, and the study analyzes emergent mechanisms such as output truncation and history compaction.

By Qiankai Xu