Paper page - SIA: Self Improving AI with Harness & Weight Updates

Source: https://huggingface.co/papers/2605.27276

Abstract

A self-improving AI framework simultaneously updates both model weights and task-specific agent architecture through a language-model feedback agent across legal classification, GPU optimization, and biological data denoising tasks.

Humans are the bottleneck in building and improving AI. Both the models and the agents that wrap them are written, tuned, and corrected by people. The long-horizon goal of an AI that can figure out how to improve itself remains open. Two largely disjoint research lines attack this bottleneck. Theharness-updateschool has a meta-agent rewrite the scaffold of atask-specific agent(its tools, prompts, retry logic, andsearch procedure) while themodel weightsare held fixed. Thetest-time trainingschool uses hand-written RL pipelines to update the model’s own weights on task feedback while the harness is held fixed. These two silos operate in isolation. We propose SIA, aself-improving loopin which alanguage-model agent(theFeedback-Agent) updates both the harness and the weights of atask-specific agent. We evaluate across three contrasting domains: Chinese legal charge classification, low-level GPU kernel optimisation, and single-cell RNA denoising. Combining both levers outperforms scaffold iteration alone on all three benchmarks. The gains are 56.6% on LawBench, 91.9% runtime reduction on GPU kernels, and 502% on denoising over the initial baseline. Harness updates make the model agentic, shaping how it searches and acts, whileweight updatesbuild thedomain intuitionthat no prompt or scaffold can instil.

View arXiv pageView PDFProject pageGitHub754Add to collection

Get this paper in your agent:

hf papers read 2605\.27276

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.27276 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.27276 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.27276 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.