Reproduce “Infrastructure for Verifiable AI: Where Proof Costs Go and How Hardware Can Help” 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

Yihan Xia, Wenxin Xu, Taotao Wang

The authors trace proof-generation costs for AI models running on external hardware, from workloads through proof systems and kernels to hardware. Measurements show bottlenecks vary by workload and prover. In VeriLoRA, fixed-weight openings stayed near 11.5 seconds for 7B and 12.4 seconds for 13B as input length grew, unlike intermediate-tensor openings; the findings apply to measured configurations, not all proof systems.

Summary by alphaXiv.

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