What the work actually involves

Turing builds evaluation and training data for frontier AI labs. On a biology project, your output is not conversation — it is a problem set. You author questions that a strong graduate student could solve with careful reasoning but that a current model plausibly gets wrong, then write the canonical solution: every inference step stated, every assumption named, units and directionality checked. Typical failure modes you are asked to target are conceptual abstraction (a model that recites the central dogma but mishandles a novel regulatory scenario), multi-step reasoning (pedigree analysis, flux through a coupled pathway, multi-generational crosses), and data interpretation (reading a gel, a growth curve, a differential expression table, or an enzyme kinetics plot).

You will also review and annotate model output — marking where a chain of reasoning goes wrong rather than only whether the final answer is right — and work with researchers to align problem types with specific benchmark goals. Contributions to new benchmark definitions across biology curricula are part of the brief, so the ability to articulate why a topic is diagnostic matters as much as knowing the topic.

What the screen looks for

  • Real depth, tested by follow-up. Screeners push past the first correct answer to ask why an alternative mechanism is wrong. Textbook recall without mechanism shows up quickly.
  • Unambiguous problem writing. A question with two defensible answers is a defective item. Expect to be probed on how you eliminate ambiguity.
  • Written English under structure. Solutions are read by researchers and other annotators; clarity and consistent notation are graded.
  • Calibration. Knowing where your own certainty ends, and flagging it, beats confident guessing.

Logistics

Fully remote contractor engagement — no medical or paid leave, no equity, no guaranteed hours. Work is asynchronous with intermittent live syncs with researchers, and volume moves with project demand; extensions depend on performance and pipeline. You need your own desktop or laptop and reliable internet. Pay is not disclosed in this listing; rates on Turing STEM projects are typically hourly or per-accepted-item and vary by degree level, specialization, and project, so treat any figure you hear secondhand as unverified until it is in your contract.