What the work actually involves

You write technical prompts in French at the level you would use with a competent colleague in your subfield — not textbook questions, but the kind of query that separates a model with real procedural knowledge from one producing plausible-sounding prose. You then read the model's response and annotate it along several axes at once: is the chemistry or biology correct, is it actually useful to the person asking, and did the model handle the dual-use dimension appropriately. Much of the judgment sits in that last axis. A refusal to explain a standard undergraduate purification is a failure; a cheerfully detailed synthesis route for something in a schedule list is a different kind of failure. You will also apply a written taxonomy to classify prompts and conversations, which means reading guidelines carefully and applying them consistently across hundreds of items rather than substituting your own instincts.

What the screen looks for

Mercor's screening is AI-led and follow-up heavy. Expect it to pick one claim from your background — a synthesis you ran, an assay you validated, a radiochemistry course — and push two or three layers deeper on reagents, conditions, failure modes, and controls. Vague answers collapse quickly. Fluency is tested in use, not asserted: you should expect to discuss your own research in French and to demonstrate that your technical French is genuinely native-register, not English terminology with French grammar around it. The safety-judgment portion is where most candidates are separated. The platform is not looking for maximal caution; it is looking for someone who can articulate why a particular piece of information is or isn't hazardous, and who understands that over-refusal degrades a model as surely as over-disclosure endangers it.

Logistics

  • Fully remote and largely asynchronous, with work drawn from a queue rather than scheduled shifts.
  • Contributors typically commit somewhere between 10 and 30 hours a week; some projects request a minimum weekly floor.
  • Start date is immediate — onboarding and calibration usually precede any paid production work.
  • Basing in France or Western Europe is preferred for time-zone overlap on calibration calls, but applicants elsewhere are explicitly welcome.
  • Pay of $58–62/hr reflects rates observed on comparable Mercor CBRN and bilingual STEM projects; actual rate depends on the project tier and is not guaranteed.