Reproduce “TCSExam: Evaluating Frontier LLMs on the Full Range of TCS Research” 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
Yifan Luo, Zhengying Liu, Yang Yuan, Zhilin Yang, Andrew Chi-Chih Yao
TCSExam evaluates four kinds of theoretical-computer-science research output: complete formal proofs, reusable solvers, verified mathematical constructions, and proofs or refutations of open conjectures. Its 100 expert-reviewed tasks use Lean or executable checkers. GPT-6 Astra completed 14 of 20 theorem tasks without supplied intermediate lemmas. Accepted conjecture resolutions adapted published work; 28 other Full tasks had no accepted complete solution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.