Contribute one original research-level mathematics problem to an advanced reasoning benchmark for a leading AI research organisation. The work is remote and open globally, with submissions in English.
About the project
Create a rigorous challenge intended to exceed the capabilities demonstrated by current frontier AI models under the programme's evaluation protocol. This is problem authoring, not proof formalisation: a machine-checked proof is not required. You will provide an exact, machine-checkable answer, a complete solution and computational materials for independent verification.
What you'll do
- Design one new, unpublished and self-contained problem in your area of expertise. Include all necessary definitions, notation, conventions and hypotheses, without relying on external references.
- Specify one exact answer in a supported machine-checkable format, such as a Python integer, SymPy expression or matrix, or tuple of integers. Approximate answers are not accepted.
- Make the answer resistant to guessing under the programme's threshold; small integers, familiar constants, simple counts and yes/no answers are generally unsuitable.
- Write a complete solution, presenting the key idea first, followed by the reasoning and required computation.
- Provide reproducible code that computes or verifies the answer in approximately 60 seconds or less on ordinary hardware, plus a reference-answer function in the required submission format.
- Test the problem using the programme's harness, including repeated independent model attempts, a statement-only guessing attack and contamination or source-identification checks.
- Submit evaluation logs and difficulty metadata, including subject area, techniques, estimated expert discovery and execution times, and whether programming is required.
Requirements
- Active mathematical research at doctoral level or beyond, such as a senior PhD student with supervisor endorsement, postdoctoral researcher, early-career faculty member or professor.
- Demonstrated research depth supported by published or ongoing work. Relevant fields include number theory, combinatorics, algebraic geometry, analysis, topology, differential geometry, PDEs, harmonic analysis, category theory and mathematical physics.
- Programming ability sufficient to compute and verify the result, preferably in Python. Available tools include NumPy, SymPy, SciPy, NetworkX, gmpy2, mpmath and galois.
- Strong mathematical rigour, including explicit assumptions and definitions and the ability to identify gaps under adversarial review.
- Publication-standard mathematical English and the ability to revise work after independent review.
- Willingness to comply with strict confidentiality and NDA requirements, including restrictions on external testing and disclosure.