What the work actually involves

You write physics problems designed to fail a model, then show your work. A typical task starts from a topic brief — say, perturbative methods in quantum mechanics, or rigid-body dynamics with non-inertial frames — and asks you to produce an item that a strong model gets wrong for an interesting reason: a hidden sign convention, a limit that has to be taken in the right order, a symbolic manipulation that looks tractable and isn't. You then write the reference solution as a clean, step-by-step derivation with stated assumptions, units carried through, and reasoning a reviewer can audit line by line. Other tasks run the opposite direction: you read a model's attempt, mark exactly where the physics goes wrong, and write an annotation that distinguishes an arithmetic slip from a conceptual failure.

The range spans early-undergraduate mechanics and E&M up through graduate-level topics — statistical mechanics, QFT, condensed matter, GR — depending on the project. Turing also references problems at the level of competitive engineering entrance exams, so some batches are dense, multi-step, closed-form items rather than research-adjacent questions. Expect to collaborate asynchronously with LLM researchers on what "hard" means for a given benchmark; the goal is calibrated difficulty, not obscurity.

What the screen looks for

  • Verifiable depth. Which subfields you can work in unsupervised, and where your knowledge thins out. Naming your real boundary scores better than claiming everything.
  • Solution hygiene. Whether you write derivations that a second physicist can check without asking you questions — assumptions declared, symbols defined, no skipped algebra at the step that matters.
  • Adversarial instinct. Whether you have a working theory of how models fail on physics, beyond "it's bad at math."
  • Written English under length. These are long-form text deliverables; the screen samples your prose directly.
  • Throughput realism. Honest hours per week, and honest time-per-problem estimates.

Logistics

Fully remote, contractor engagement — no medical or paid leave, no guaranteed hours. Work is asynchronous and batch-based; you pull tasks and return them against a deadline rather than sitting in meetings. Current Master's, PhD, and postdoc candidates are explicitly encouraged, and the arrangement suits people who can carve out predictable evening or weekend blocks. Contract extension depends on quality scores and whether the client project continues. You need your own desktop or laptop and reliable internet. Pay for this listing is undisclosed by the platform; treat any figure you see quoted elsewhere as unverified.