Mercor keeps a standing generalist pool for AI training and evaluation projects that need careful reading and general reasoning rather than a specific credential. You write prompts and reference answers, grade model output for correctness and instruction-following, and explain in writing exactly why an answer fails.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work involves

Projects in this field fall into three rough shapes. The first is authoring: writing prompts on everyday questions that have a verifiable right answer, plus a reference answer that spells out the context a good response has to account for. The second is grading: reading model output and judging whether it is correct, whether it followed the instruction actually given rather than a nearby one, and whether it is complete without padding. The third is failure-spotting: flagging confident wrongness, self-contradiction within a single answer, and answers to questions nobody asked.

Most of the labour is written justification. A score with no reasoning is close to useless to the research team receiving it, so expect to spend real time writing a paragraph that names the specific sentence or claim that broke the answer. Task instructions and rubrics are often imperfect, and flagging an unclear rubric is treated as part of the job rather than a complaint.

What the screen is actually testing

Mercor verifies your background and runs a short AI interview, roughly 20 minutes. For generalists there is no credential gate, which means the interview leans harder on demonstrated care: can you describe your professional or academic work concretely, can you hold up under a follow-up that pushes on a claim you just made, and can you articulate a standard for "wrong" that goes beyond personal taste. Vagueness is the main way candidates lose ground here, because vagueness is exactly the failure mode the role exists to catch.

Logistics

Fully remote and asynchronous; you confirm where you work from at application.
Listings in this field have posted at $35–40/hr, set per project by scope and depth — observed rates, not a guarantee.
This is a standing pool, not an opening. It issues no accept or reject decision; you are contacted when a project matches, which may be within a week or several months.
Hours are project-defined. Actual rate, hours, and client are named on the specific project listing you get invited to, and the hiring decision happens there.
One signup covers every network you qualify for; joining more of them improves match odds.

Generalist Expert

What the work involves

What the screen is actually testing

Logistics

Interview prep — 10 questions