Reproduce “Recursive Harness Self-Improvement for Frontier Reasoning Data Synthesis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenlong Zhang, Zhengbo Jiao, Chenxu Zhang, Lekang Jiang, SiYuan Ma, Qituan Zhang, Guo Chen, Linfeng Zhang
The authors develop a recursive system that evolves reasoning problems alongside the harness used to construct them, including reusable skills, prompts, and workflow. Solver failures on valid questions inform updates during generation, while batch-level feedback can revise the harness. Across 14 rounds, DeepSeek-V4-Pro accuracy fell from 100.0% to 54.8%, indicating harder tasks for that solver and measure, not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.