Newest papers

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

Jens Lehmann, Sahar Vahdati · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30499 · cited by 0 works (OpenAlex) · open-access PDF

Arabic

ويبدو أن نظام التفتيش الذكي بشكل عام يكتسب المهارات الجديدة بشكل فعال دون أن يتطلب من البشر إعادة تصميم نظام التفتيش الذكي لكل مهمة جديدة. ومع ذلك، فإن النماذج اللغوية غالباً ما تعمل وحدها. ويبدو أن نظام التفتيش هو برنامج نموذجي يعتمد على التفتيش على النماذج، وتتبع الذاكرة، واستخدام الأدوات، وإعادة التفتيش، وإعادة توزيع التفتيش، وإعادة توزيع التفتيش، وإعادة توزيع التفتيش، وإعادة توزيع التفتيش، وإعادة توزيع التفتيش عن طريق النماذج. ونحن نسأل ما إذا كان النماذج الحدودية يمكن أن تُمكّن من بناء وتطوير مثل هذه التفتيش على نفسها. من خلال نموذج الت

Machine translation (Meta M2M100); the English original follows.

Abstract

A generally intelligent system acquires new skills efficiently without requiring humans to redesign its software scaffold for every new task. Yet language models rarely act alone. A harness is a symbolic program that orchestrates calls to the model, handling perception, memory, tool use, recovery, and inference allocation on the model's behalf. ARC-AGI-3 is an interactive benchmark in which agents must explore unfamiliar environments, figure out their goals on the fly, and keep learning as they go. On it, human-steered systems already report ceiling scores across the public games, so we know effective designs exist. We ask whether frontier models can build and improve such a harness on their own. Through a fixed coding bootstrap, each model constructs a complete task harness, acts through that harness on five development games, and revises the entire codebase twice. Every version is evaluated on twenty games that never influence adaptation. We measure task performance with Relative Human Action Efficiency (RHAE), which combines level completion with action efficiency relative to human baselines, and record dollar-denominated inference cost as a practical resource measure. Across 27 lineages, harness performance ranges from zero to near-ceiling scores. Two revisions raise mean held-out RHAE by only 2.80 points (95% interval [-1.44,7.35]). Development gains overstate held-out improvement, and reductions in inference cost can come at the expense of task performance. Human-steered systems demonstrate further headroom. These findings reveal a gap between autonomous harness construction and reliable adaptation under this two-revision protocol: even strong task performance does not ensure that subsequent revisions generalize or preserve existing capabilities.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Task (project management), Inference, Coding (social sciences), Risk analysis (engineering), Adaptation (eye)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange