About we-solve

Open problems. LLM–human collaboration. A shared effort to push the frontier of research.

we-solve brings people and large language models together to work on open research problems. The idea is simple: make it easy to find a question, learn from earlier attempts, contribute something useful and let others build on it.

A full solution is only one way to contribute. A partial result or a careful review can move research forward.

Use your top LLMs to push the frontier

If you have access to a top LLM, you can put some of that access towards an open problem. Copy a research prompt from a problem page into your usual assistant, explore an approach or examine an existing argument, then submit the work through the contribution form. A coding agent can also help prepare a submission through GitHub.

You use your own account and choose how much time and model usage to contribute. You share the resulting research; we do not ask for your login or API keys. Contributions made without an LLM are equally welcome.

See how to contribute

A shared research database, powered by GitHub

GitHub provides the public record: submissions, revisions, attribution and discussion. Published attempts and reviews are linked to their problems, so the next contributor can see what has been tried, what was checked and what remains unresolved.

We’re building a workflow that combines automated format checks, model-assisted screening and community review for both human and LLM contributions. Models help flag issues and assess whether a submission’s claims match its evidence. Contributions that meet the publication requirements can then join the searchable research database.

Problem sources

For Erdős problems, we track problems from two key resources:

Terry Tao’s Erdős Problems Database: a collaborative database of Erdős problems with progress tracking.
erdosproblems.com, by Thomas Bloom: detailed problem statements, historical context and relevant comments.

Rankings and problem identity

In every field, a problem’s rank is separate from its permanent, unique identity. Displayed labels such as Math-1, CS-1 and Econ-1 show the current rank within a field. Permanent problem IDs and imported source IDs remain internal and do not change when ranks change. A problem listed in more than one field can have different ranks while keeping the same research record and contribution history.

Rankings are provisional estimates of difficulty. See the ranking policy for the scope and limits of the current assessments.

Lean contributions

Lean contributions use a short TeX description with an external link to the complete Lean source and original generated output. They appear alongside other attempts on the relevant problem page. The description states the formalised scope, remaining gaps, model attribution, Lean/Mathlib versions and actual checking outcome.

OpenAI results

OpenAI entries identify summaries of claimed results published in OpenAI’s math repository. A recorded full-solution claim concerns the complete question on the problem page; a special case, weaker bound or result covering only part of the question is a partial claim. These remain attributed claims and have not been independently verified here. The original source catalogue status is shown separately, and community reviews record the scope and evidence of any checking.

Records include the claimed mathematical hypotheses and conclusion, with links to the source manuscripts and any available Lean scope and Comparator files at a fixed revision. A claim attributed to OpenAI does not change the original source status.

Economics and the social sciences

Unlike mathematics and certain areas of theoretical computer science, where major open questions—such as the Millennium Prize Problems—can be precisely formulated and their solutions may fundamentally transform their fields, economics has no comparably established set of mathematically well-defined problems whose resolution would necessarily reshape the discipline. Many of the most important open questions in economics, political science and the social sciences more broadly are conceptual, methodological or empirical rather than purely mathematical.

The economics problems listed here were ranked primarily according to the estimated difficulty of solving them, as assessed by GPT-6 Astra Pro, rather than their prominence, practical significance or potential impact on the discipline. The ranking also gives some priority to particularly difficult problems within individual subfields.

Nevertheless, certain problems in economics could have transformative implications if resolved in the affirmative, even when such a resolution appears highly unlikely. Some of these are included in the catalogue.

we-solve

This website was designed and created through LLM–human collaboration, with Mehmet Mars Seven of King’s College London as the human collaborator and maintainer of this site.

It is a new and more structured successor to the Erdős Problems LLM Hunter repository, designed to make contributing easier. That repository was forked from Terry Tao’s database.

Acknowledgments

Paata Ivanisvili, for the initial idea of attacking MathOverflow problems with LLMs.
Terry Tao, for the Erdős Problems Database.
Thomas Bloom, for erdosproblems.com.
Ulam AI and the UnsolvedMath Contributors, for the UnsolvedMath database.

Disclaimer

we-solve is an open research workspace and archive, not a peer-reviewed journal. Human and LLM contributions may contain errors, invalid proofs, invented references or false claims. Passing automated checks or model screening, appearing in the database, or receiving a favourable review does not establish correctness, novelty or validity.

Read claims critically, check sources and reproduce results wherever possible. Independent scrutiny and validation are essential. Promising work should be developed into papers for proper peer review.

Give full credit to people and tools, cite sources accurately, and respect licences and third-party rights. A claim made here does not change a problem’s official source status.

Review and screening · Attribution · Licences and third-party notices