Reproduce “Input-Conditioned Search for Efficient Test-Time Scaling of Recursive Reasoners” 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

Haochen Shi, Tianwen Fu, Qianggang Ding, Yinhan Lu, Yizhan Li, Huan Zhang, Yajie Zhao, Xingdi Yuan and 2 more

ICS scales puzzle solvers at test time by changing the input condition while keeping weights frozen and inducing a different update map from the initial state. It searches hypotheses rather than repeating or perturbing one map. On Sudoku-Extreme, it reaches 99.70% accuracy at 1.14 TFLOPs per puzzle, but evidence is limited to structured puzzles and does not guarantee recovery from failure.

Summary by alphaXiv.

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