← Back to jobs

AI training & evaluation

STEM Expert Task Pilot

Mercor

Build one hard, real-world task in your field, used to test how well AI agents handle expert-level technical work.

Work arrangement & location
Remote
Compensation
$2,000 per task
Engagement
task-based
  • Life, Physical, and Social Science

Employer description and requirements

Build one hard, real-world task in your field, used to test how well AI agents handle expert-level technical work.

Each task is open-ended: there is no single right answer, and an automatic scorer grades the AI agent's result from 0 to 1 with partial credit. For example: "Build the fastest possible attention engine."

  • Pay: $2,000 per task, paid once your finished task is accepted
  • Time: about 20 hours
  • Apply by: Oct 4, 2026, 11:59pm PT
  • Task due: Oct 6, 2026, 9pm PT

To apply, pick the task types that fit you and describe one task idea in a few paragraphs. You don't build anything until your idea is approved. We review ideas within 12 hours.