What the work actually involves

You write problems designed to fail a model, then you write the solution that proves it failed. Most days consist of authoring original physics and STEM items — not pulling them from textbooks — targeting known weak spots: multi-step symbolic manipulation, dimensional reasoning, abstraction across formalisms, problems where a plausible-looking shortcut gives the wrong answer. Each item ships with a fully worked, step-by-step solution in clear English, with the reasoning articulated rather than compressed into algebra. A second strand of the work is reviewing and annotating model outputs: identifying precisely where a chain of reasoning breaks, not merely that the final number is wrong.

You will also be asked to help shape benchmark coverage — mapping a physics curriculum (classical mechanics, E&M, statistical mechanics, quantum, and beyond) onto difficulty tiers, and flagging where existing item pools are thin. That means working with LLM researchers who know evaluation design but not your subfield, so you need to explain why a particular problem is hard in terms they can act on.

What the platform screens for

  • Real graduate-level depth. Screens probe a declared specialism with follow-ups; surface familiarity collapses quickly.
  • Problem authorship, not problem solving. Can you construct an item with a single defensible answer, no ambiguity in the setup, and a genuine difficulty source?
  • Written clarity. Solutions are the product. Terse derivations that skip justification get rejected in review.
  • Error localisation. Given a wrong model answer, can you name the exact step and the exact misconception?
  • Freelance reliability. Realistic weekly hours, responsiveness in async threads, and a working desktop setup.

Logistics

Fully remote, contractor engagement — no medical or paid leave, no guaranteed hours. Work is typically async with periodic calibration sessions or written feedback rounds from reviewers; extensions depend on throughput and on whether your items survive quality review. Turing does not publish a rate for this listing; pay observed on comparable Turing STEM authoring work varies by degree level, specialism scarcity, and project, and is negotiated at offer. Expect an initial paid or unpaid sample task before assignment.