SIA: Self Improving AI with Harness & Weight Updates
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917542045417472 |
|---|---|
| author | Hebbar, Prannay Manawat, Yogendra Verboomen, Samuel Ivanova, Alesia Palanimalai, Selvam Bhatia, Kunal Baskaran, Vignesh |
| author_facet | Hebbar, Prannay Manawat, Yogendra Verboomen, Samuel Ivanova, Alesia Palanimalai, Selvam Bhatia, Kunal Baskaran, Vignesh |
| contents | 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. The harness-update school has a meta-agent rewrite the scaffold of a task-specific agent (its tools, prompts, retry logic, and search procedure) while the model weights are held fixed. The test-time training school 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, a self-improving loop in which a language-model agent (the Feedback-Agent) updates both the harness and the weights of a task-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. SIA-W+H achieves 25.1% over prior SOTA on LawBench, 12.4% faster GPU kernels than prior SOTA (1,017 vs 1,161 μs), and 20.4% over prior SOTA on denoising. Harness updates make the model agentic, shaping how it searches and acts, while weight updates build the domain intuition that no prompt or scaffold can instil. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_27276 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SIA: Self Improving AI with Harness & Weight Updates Hebbar, Prannay Manawat, Yogendra Verboomen, Samuel Ivanova, Alesia Palanimalai, Selvam Bhatia, Kunal Baskaran, Vignesh Artificial Intelligence Computation and Language I.2.6; I.2.11; I.2.8 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. The harness-update school has a meta-agent rewrite the scaffold of a task-specific agent (its tools, prompts, retry logic, and search procedure) while the model weights are held fixed. The test-time training school 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, a self-improving loop in which a language-model agent (the Feedback-Agent) updates both the harness and the weights of a task-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. SIA-W+H achieves 25.1% over prior SOTA on LawBench, 12.4% faster GPU kernels than prior SOTA (1,017 vs 1,161 μs), and 20.4% over prior SOTA on denoising. Harness updates make the model agentic, shaping how it searches and acts, while weight updates build the domain intuition that no prompt or scaffold can instil. |
| title | SIA: Self Improving AI with Harness & Weight Updates |
| topic | Artificial Intelligence Computation and Language I.2.6; I.2.11; I.2.8 |
| url | https://arxiv.org/abs/2605.27276 |