About Future of CS Theory: Method, AI Use and Conflicts

What this site is

Future of CS Theory is a privately run ledger of results in theoretical computer science that are credited to AI, and of how far each one has been publicly checked. It is not affiliated with OpenAI, Anthropic, Google or any researcher named here. Its one job is to keep separate things separate: a claim, a formalization, a person's statement and a peer-reviewed publication.

The boxes on the ledger borrow the mark that ends a mathematical proof. Most of them are still hollow.

How a row gets in

• A result is listed when a primary source, such as a paper or an official repository, makes the claim and credits AI with a substantial part of the work.
• Bounds and theorem statements are copied as written and never normalized.
• Every cell links the source it came from. If we found nothing, the cell says so and gives the date.
• Named people are paraphrased, with date and link. If someone said they had not checked a result, that is what we report.
• A filled box means a public record exists. It says nothing about whether the mathematics is right.
• A half-filled Lean box means the repository or its documents say coverage is narrower than the paper's main theorem. A full box means the repository lists the main statement as formalized. We have not audited that coverage ourselves.
• Each row carries a last-checked date, and changes go in the changelog.

The first version covers 12 results: ten prominent in the October 2026 discussion and two earlier AlphaEvolve results from a 2025 paper. It is not a complete census. OpenAI's repository alone lists hundreds of manuscripts.

What we did not do

We did not review any proof, run Lean or Comparator, or reproduce any algorithm. No theoretical computer scientist has reviewed this site. "None found" means our searches found nothing. They covered the repositories, arXiv metadata and the web, not every proceedings volume, mailing list or private exchange. Some social posts were read through public API mirrors rather than the original page.

AI use and conflict of interest

This site was researched and drafted with Claude, a model made by Anthropic. A person chose the scope and decides what is published. A second pass by the same Anthropic model rechecked the cells against their sources. It was not a person, it is not independent of Anthropic, and it could not reach every source.

Anthropic is a party to events covered here. It published one of the two ETH claims, its model is credited with the 3SUM and APSP algorithms, and it is a funder in a Simons Foundation program for mathematicians. Treat comparisons between labs with that in mind, and tell us if you see a slant. The site takes no money from AI companies and runs no ads.

Corrections

If an entry is wrong, or a statement of yours is misreported, tell us with a source. Fixes are dated in the changelog.

Credits

Headings are set in CMU Serif, a version of Donald Knuth's Computer Modern, under the SIL Open Font License. Body text is set in Inter.