Reproduce “Beyond the Training Distribution: A Survey of Reasoning Generalization in Large Language Models” 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
Tue Le, Van Dong Le, Hai Dang Luu, Bui Duc Duan, Hoang Tran Vuong, Huy Huu Nguyen, Linh Ngo Van, Ngoc-Thanh Dinh and 3 more
The authors survey how large language models generalize reasoning procedures across explicit distribution shifts. They organize methods by whether they act during training, inference, or architecture, and review behavioral, mechanistic, and theoretical evidence. The survey finds that methods can help in particular settings, but none consistently generalizes across shifts; differences among studies preclude controlled comparisons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.