The jobs exam

Benchmarks usually ask one AI to mark another. This one asks people who do the job. A working payroll officer, nurse, paralegal or site manager writes a real task and the marking scheme they'd use on a junior colleague. AI answers are then graded blind by others in the same occupation. A result counts only when two or more of them have marked it.

Three ways to take part
  1. Write a paper (20 minutes): a task from your work, and how you'd mark it. Submit a paper.
  2. Grade answers in your occupation, below. You won't see which AI wrote what.
  3. Run models on the papers, free with local models: skillbench/jobs-exam.

Loading the papers…