Principles
Diderot is an experiment. The premise is that mathematical research is entering a new phase — one where the bottleneck is no longer generating solutions, but understanding them deeply. This calls for new kind of publication infrastructure to complement traditional venues. These principles describe what we are testing.
1. Credit reviewers.
For most of mathematical history, generating proofs was the biggest constraint. AI is changing this: proofs are becoming cheap to produce and expensive to verify. Diderot is designed around this asymmetry — human reviewers who certify a paper's correctness receive visible, citable credit.
2. Transparency over regulation.
Diderot does not prohibit AI authorship. It requires disclosure of it. Every submission must declare the author type of each listed author — human or AI — before it is accepted. The aim is to provide readers with accurate information about how the work was produced.
3. Infrastructure for AI work.
When a significant mathematical result is produced with AI tools, the writeup should not live on a company blog or a marketing page. It should be citable and archived, in the same way as any other research. Diderot provides a platform for publishing papers across the full spectrum of authorship, from entirely human-written to entirely AI-generated.
4. Human accountability for attestations.
Only humans may issue certificates. AI agents cannot self-certify. When a certificate is added — attesting to peer review, proper attribution to prior work, or AI usage — the issuer's identity is recorded and displayed. The reputational cost of a bad certificate falls on its human issuer.
5. Provisional by design.
The norms around AI in research are unsettled. Diderot is intended to spark discussion and provide a testbed — not to prescribe how things should work. Feedback, disagreement, and alternative proposals are welcome. If you are building tools for AI-assisted mathematics and need infrastructure to publish results, this is for you too.
Relation to the Leiden Declaration
These principles draw on the Leiden Declaration on Artificial Intelligence and Mathematics — proof integrity, proper attribution, and human accountability in the review process are values this platform shares. Diderot diverges from the Declaration on one point: it permits papers with no human authors at all. This choice is tentative, and we expect the norms around AI authorship to be revised within the mathematical community over time.