What the work actually involves
You are producing training and evaluation data for frontier language models in physics. In practice that means three recurring tasks: authoring original problems at or above qualifying-exam difficulty that a strong model still fails; writing complete, verifiable solutions with every step of the derivation shown; and reviewing model outputs to say precisely where a chain of reasoning goes wrong — a dropped factor of 2π, a boundary condition applied in the wrong limit, a dimensionally consistent answer built on a physically impossible assumption.
Most tasks come with a rubric and a subfield tag (classical mechanics, E&M, quantum, statistical mechanics, condensed matter, particle/high-energy, astrophysics, computational). Work is submitted through the platform's task interface, usually in LaTeX. Reviewers send items back when a solution is correct but under-justified, so the standard is not 'I know this is right' — it is 'a competent physicist reading only what I wrote can confirm it is right.'
What the platform screens for
micro1 runs an AI-led interview and, for technical categories, a domain assessment. The screen is looking for confirmed graduate-level training (PhD in progress, PhD, or postdoc), a nameable specialisation you can go deep on under follow-up, and the ability to explain a derivation cleanly in writing. Publication record and teaching or grading experience help. Vague breadth hurts — a candidate who claims all of physics and cannot handle a pointed question in any of it screens out fast.
Logistics
- Fully remote, contractor basis, async — no fixed shifts
- Typical commitments run 10–20 hrs/week, though some contributors work more when project volume allows
- Work is project-based and can pause or ramp between engagements
- LaTeX fluency is assumed; time zone is generally flexible
- Pay bands are as observed from contributor reports, not guaranteed by the platform