What the work actually involves

You write expert-level prompts in Netherlands Dutch across your scientific specialism, then read what the model gives back and judge it. The judgment is layered: is the chemistry or biology correct at the level a working researcher would demand, is the answer actually useful to someone asking in good faith, and does it handle sensitive or dual-use material responsibly. A third strand of the work is classification — applying a written guideline document to sort prompts and conversations into categories, consistently, across many items. That last part is less glamorous than it sounds and it is where most disagreement between annotators shows up.

The safety dimension is central rather than incidental. Coverage spans chemical, biological, and radiological/nuclear subject areas, which means you will regularly be near information that is legitimately taught in graduate programmes and also legitimately restricted. The work is not to invent attacks; it is to test whether a model draws the line where a competent scientist with security awareness would draw it, and to explain in writing why a given response sits on the wrong side of it.

What the screen is looking for

  • Genuine PhD-level depth in one subfield, ongoing or completed, in chemistry, biology, or an adjacent discipline — the screen probes downward into synthesis, molecular biology, virology, radiochemistry, or whatever you claim, and generalists tend to run out of substance by the second follow-up.
  • Netherlands Dutch, not Dutch generally. A parallel role covers Belgian Dutch. Expect to be asked about terminology and register as used in Dutch universities and labs, and to write technical prose in Dutch without falling into English calques.
  • Business-level written English, since guidelines, rubrics, and reviewer feedback run in English.
  • Current residence in the Netherlands — this is a stated gate, not a preference.
  • Calibrated dual-use judgment. Both reflexive refusal of everything hazardous and casual willingness to spell out procedure are failure modes.

Logistics

Remote and asynchronous, work drawn from a queue rather than scheduled shifts. Start date is immediate. Contributors on Mercor projects of this type typically commit somewhere between ten and thirty hours a week, and the screen will ask what you can realistically sustain alongside a lab position or dissertation — an honest smaller number reads better than an inflated one you then miss. Pay is observed at $61–65/hr for this listing; rates on the platform vary by project and are not guaranteed.