$80–110/hr
The role in one line
A physics PhD produces and evaluates research-level problems to benchmark AI models on genuine scientific reasoning.
Written by Training Turk from the public listing; it may be incomplete or out of date. Read the full posting on Mercor.
What you would do
- Generate or solve frontier-level physics problems tailored to your specific publications and expertise
- Write solutions and reasoning that meet academic proof standards for peer verification
- Review submissions from other physicists and audit work quality
- Create problems designed specifically to test whether AI can handle real research reasoning
- Collaborate to match your expertise with one of four specialized mathematical physics domains
Who they are looking for
- Doctorate in mathematical physics, theoretical physics, or mathematics (required)
- Published first-author work on your chosen phenomenon within the past five years
- Proficiency with LaTeX, Python, SymPy, and Jupyter notebooks
- Documented knowledge of one or more specialized subfield method families
- English at B2+ level with ability to write clear scientific reasoning
Skills this role asks for
What the interview is likely to probe
1.Specialized domain depth
Confirming your research sits within one of four narrow specialties, not just general mathematical physics, ensures work quality and alignment with project scope.
Expect something like: “Your dissertation focused on hypergeometric identities, but recent papers explore asymptotic expansions. How does your hypergeometric work apply to Gaussian hypergeometric functions and parameter differentiation?”
2.Proof-level clarity
Solutions must be verifiable by peer specialists without requiring author clarification, a standard stricter than most physics papers.
Expect something like: “You have authored a solution for a Lax pair problem. A colleague in your subfield reads it cold. What specific details would they need to independently confirm your analytic continuation steps?”
3.Original problem construction
Creating problems that accurately test AI reasoning, not too easy and not trivial textbook variants, requires understanding both the domain and genuine difficulty.
Expect something like: “Design a problem on Calogero-Moser systems that an AI could attempt but would miss if it merely memorized textbook conserved charge formulas. What feature makes it genuinely difficult?”
4.Rapid publication verification
Your papers are checked against arXiv and journal databases before matching; early clarity accelerates onboarding.
Expect something like: “In your application you list five papers. Two are on conformal geometry, three on special functions. The conformal papers are 2023 and 2021. How do you expect us to verify your first-author status and the methods you claim for each?”
How to prepare
- Review the CritPt benchmark paper (arXiv:2509.26574) and confirm your published work aligns with one of the four listed specialties
- Gather three to five first-author papers and prepare to map their methods to the required technique families
- Test your proficiency with LaTeX, Python, SymPy, and Jupyter by solving a sample multi-step special function or integrable systems problem
- Prepare an explanation of how your subfield work differs from related but excluded areas (e.g., why your Haldane-Shastry papers fit but not adjacent topics)
Facts
- Pay
- $80–110/hr
- Commitment
- hourly
- Hours
- 10 per week
- Work arrangement
- remote · Remote
- Domain
- Life, Physical, and Social Science
- Posted
- 9/25/2026
- Open slots
- 3