The research teams behind the best-known AI models come to Mercor for scientific judgment they can't generate on their own. Your standards and expertise become the rubric that frontier AI is graded against.
About the role
The research teams behind the best-known AI models come to Mercor for scientific judgment they can't generate on their own. Your standards and expertise become the rubric that frontier AI is graded against.
This is a standing listing for researchers who want AI training and evaluation work, not a specific job opening. It stays open whether or not a matching project is running this week.
New projects start on short notice, so instead of posting and pulling a listing each time, we keep one open per field and draw from it when one begins.
What the work looks like
We match researchers in the network to projects like:
- Writing a research problem from your own field, with the system, the data, and the assumptions, plus the reference answer a model's attempt gets graded against
- Grading model-written analysis and interpretation: whether the method suits the question, whether the statistics support the claim, whether the conclusion follows from the evidence
- Flagging where a model produces a result that is internally consistent and would not survive peer review, given the design, the sample, or what the measurement can actually show
Projects across these fields have posted at $60 to $120 an hour, set per project by scope and depth.
What we're looking for
- A graduate degree in a life, physical, or social science, with research experience
- Clear written communication, since most of the work involves explaining your reasoning
- Comfort with ambiguous tasks, and willingness to flag unclear instructions
What this covers
This pool spans the sciences, including:
- Biology, biotechnology, chemistry, and materials science
- Physics and astronomy; mathematics
- Earth and environmental science; agriculture
- Psychology; sociology and anthropology
- Economics; political science and policy research
Where you trained matters less than whether you designed the studies and interpreted the data yourself, recently enough to explain the calls you made.
How it works …
Official source: work.mercor.com. Details can change; the platform's page is the final word. How we check listings.
