Reproduce “Human agency in initial human-AI proof formalization workflows” 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

Katherine M. Collins, Simon Frieder, Jonas Bayer, Jacob Loader, Jeck Lim, Peiyang Song, Fabian Zaiser, Lexin Zhou and 9 more

A study characterizing initial human-AI proof formalization workflows revealed that users prioritize high-level human control while adapting multi-tool AI approaches, resulting in improved formalization accuracy without a significant change in completion time.

Key insights

  • Mathematicians strongly prefer to retain high-level creative and strategic control over the formalization process, even when leveraging AI assistance.
  • Users actively construct flexible, multi-tool AI workflows, deliberately selecting and combining different AI tools to compensate for individual tool limitations.
  • Despite frequently cited frustrations with AI tool reliability, even current imperfect AI tools can lead to improved formalization accuracy in practice.

Summary by alphaXiv.

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