Mercor keeps a standing physics pool that frontier AI labs draw from when they need derivations written, graded, and stress-tested by someone who has actually done the work. You write problems from your own subfield with reference solutions, and grade model attempts on whether the approximations, limits, units, and scaling hold up — not just whether the final expression matches.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work actually involves

Projects vary, but they cluster around three tasks. The first is authoring: you write a problem from your own research area — the system, the stated assumptions, the regime of validity — and supply the reference solution a model's attempt will be graded against. The second is grading model-written derivations and numerical work, which means checking whether the approximation invoked is legitimate in the regime claimed, whether the limiting cases recover known results, and whether dimensions and scaling survive each step. The third is the subtle one: flagging where a model lands on the correct final expression through reasoning that would collapse if a single assumption were changed. Most of your output is written prose explaining why a step is wrong, not a score.

What the platform screens for

Mercor verifies a graduate degree in physics or a closely related field plus research experience, and the AI interview probes whether you personally did the derivations and analysis recently enough to defend the judgment calls. Expect follow-ups that go a layer deeper than your first answer — naming a subfield is not enough; you will be asked about the specific approximation you relied on and where it breaks. Written clarity carries real weight here, because the deliverable is an explanation. Comfort with underspecified tasks matters too: projects often arrive with ambiguous rubrics, and flagging the ambiguity is treated as correct behaviour rather than as a complaint.

Logistics and honest expectations

This is a standing listing, not an opening. It does not accept or reject anyone — you apply once, complete a roughly 20-minute AI interview, get your background verified, and join the physics pool. When a client project needs physics, matching people are invited to a specific listing that names rate, hours, and client, and the hiring decision happens there. The wait can be a week or several months depending on what labs are running.

Fully remote and asynchronous; work is done against deadlines, not fixed shifts
Past rates have run $70–100/hr, set per project by scope and depth — observed, not guaranteed
Hours are typically part-time and project-bound, layered alongside academic or industry work
Joining additional networks you qualify for improves match odds; one signup covers all of them

Physics Expert

What the work actually involves

What the platform screens for

Logistics and honest expectations

Interview prep — 10 questions