What the work actually involves
You write scientific prompts in Polish at the level of your own research, then judge how a frontier model answers them. Some tasks ask for a technically demanding question in your subfield — a synthesis route, an assay design, a decay-chain calculation — and an assessment of whether the model's response is accurate, useful, and correctly scoped. Other tasks are classification: applying a written rubric to decide what category a prompt or conversation falls into, and doing it consistently across dozens of items. A meaningful share of the material sits near dual-use territory, so you will regularly have to distinguish a legitimate technical question from one where a fully detailed answer would be irresponsible, and articulate that line in writing rather than just sensing it.
What the screen looks for
The platform interview is AI-led and probes three things. First, that the doctorate is real and specific: expect follow-ups on your actual thesis chemistry or biology, instrumentation you have personally operated, and where your knowledge stops. Second, that your Polish is genuinely native-or-near-native for technical registers — terminology, not conversational ease — alongside written English good enough to follow guidelines and file rationales. Third, safety judgment: whether you can reason about hazard, uplift, and publicly available information without collapsing into either "never answer" or "information should be free." Depth in one subfield beats a survey of many; candidates who claim CBRN breadth they cannot support tend to fall apart under the second follow-up.
Logistics
- Fully remote and asynchronous; work is claimed from a task queue on your own schedule.
- Roughly 7 hours per week, part-time, starting immediately.
- Poland or the wider Eastern Europe region is preferred but not required.
- Observed pay of $50–54/hr as posted by the platform — bands shift by cohort and are never guaranteed.
- Training on the annotation workflow is provided; no AI/ML experience needed.