Reproduce “Expanding LLM Reasoning” 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
Rian Atri, Evan Luo
The authors test whether reasoning models can allocate inference by preserving a chain’s prefix and regenerating its suffix from a stored step, rather than sampling only complete chains. They measure each restart’s expansion utility against the original answer and compare placement rules with self-consistency. On DeepSeek-R1-Distill-Qwen-14B/MATH-500, always-last reaches 0.830 accuracy, 0.052 above the matched-output frontier in one 100-problem panel.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.