Reproduce “Asking the Right Questions: Improving Reasoning with Generated Stepping Stones” 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

Hengyuan Hu, Tingchen Fu, Minqi Jiang, Alexander H Miller, Yoram Bachrach, Jakob Nicolaus Foerster

ARQ adds a generated, solved subproblem to a reasoning model’s attempt at a target problem, using both the subproblem and its solution as context. The authors train stone generators with downstream-success scores and report average target-accuracy gains of 3.1% for Qwen3-8B and 2.9% for Qwen2.5-32B across three math datasets; these results do not establish broader gains or test-time efficiency.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.