For physicists
Frontier AI models are being pushed hard into physical reasoning, and their errors in this domain are subtle enough that only trained physicists reliably catch them. That is where the current work is.
Physical reasoning is one of the areas where models have improved most dramatically in the last two years and also one where their remaining errors are most costly. A model that can solve an undergraduate mechanics problem cleanly might still confidently produce a gravitational-lensing scenario that violates conservation of mass, or misapply thermodynamic reasoning at the boundary between classical and quantum regimes. Catching those errors takes trained physicists. Frontier labs know this, and the expert-annotation programs for physics have expanded accordingly.
What the work looks like
- Response evaluation. You review model outputs on prompts across mechanics, electromagnetism, quantum theory, statistical mechanics, general relativity, condensed matter, particle physics, or astrophysics. You score them against what a working physicist would say and correct where the reasoning fails.
- Problem authoring. You write problems at graduate or research level with clean solutions, dense enough for training use.
- Adversarial testing. You construct problems where the model is likely to blend related but distinct regimes, misapply a limit, or hallucinate a plausible-sounding derivation.
- Rubric design. You author the criteria a physicist would use to grade a derivation or a problem attempt, and audit the criteria the lab is currently using.
The work is asynchronous. Sessions run in blocks of two to four hours. Structured deliverables.
What labs actually look for
Graduate training in physics, astrophysics, or a closely related field. Active engagement, whether through research, teaching, or serious independent work. Written English at the level of a physical review paper.
They do not look for machine learning experience. Almost no physicist hired for this work has ever trained a model.
Where the pay lands
Rates are in USD. For physicists in emerging markets, this typically exceeds a postdoctoral stipend or a lecturer salary on a per-hour basis by a large multiple. For physicists in higher-cost markets, it is competitive with technical consulting.
How to enter
Sign up, upload a recent CV that shows your sub-field, degrees, and current activity, and let the platform surface the roles currently open in physics annotation, evaluation, and problem authoring.
See roles that fit you.
Sign up, upload your resume, and browse the roles that actually match your training. Free for candidates.