What the work involves
You write expert-level prompts in Korean — the kind of question only someone who has actually run the experiment would ask — and then evaluate what the model returns. Days split between prompt authoring in your subfield, annotating responses for scientific accuracy and helpfulness, and applying a structured taxonomy to classify prompts and conversations. A meaningful share of the work sits on the safety boundary: distinguishing a legitimate synthesis or protocol question from one that is functionally a request for uplift, and judging whether a refusal was correct, over-broad, or a false comfort that leaked the answer anyway.
Subject coverage spans chemical, biological, and radiological/nuclear areas. Chemists coming from radiochemistry, nuclear chemistry, energetic materials, or radiation physics are explicitly in scope. No prior AI/ML background is expected — the workflow and the guidelines are taught. What is not taught is the domain depth or the Korean.
What the screen looks for
- Verifiable PhD depth in one subfield, not survey-level breadth. Expect follow-ups that push past the first answer: reagents, failure modes, instrumentation, why a published protocol behaves badly in practice.
- Genuine Korean scientific register. Not conversational fluency — the ability to write technical Korean the way a Korean-language lab group or journal actually writes it, including where English loanwords and Sino-Korean terms are conventional.
- Dual-use judgment under ambiguity, reasoned rather than reflexive. Blanket-refusing everything hazardous scores as poorly as being naive about it.
- Guideline discipline — whether you can apply someone else's rubric faithfully when you privately disagree with it.
Logistics
Fully remote and largely asynchronous, with contributor-set hours and per-hour compensation observed in the $63–67 range for this posting; rates are set by the platform and vary by track and assessment outcome. Start is immediate. Based in South Korea or East Asia is a stated preference only — applicants elsewhere are welcome. Volume on Mercor projects fluctuates, so treat this as flexible contract work rather than a guaranteed weekly load, and be honest in the screen about the hours you can actually commit.