What the work actually involves

You write problems that a strong AI model should fail. Each task starts from real research artifacts — a raw dataset, a published protocol with a subtle flaw, a supplementary table, a methods section that doesn't match its figures — and asks an AI agent to do something a competent postdoc would do: spot the confound, recompute the effect, judge whether the conclusion is supported by the data. You then document the correct methodological reasoning and the correct biological interpretation, and build a multi-item rubric that an LLM grader can apply consistently: separate line items for methodological rigor, data interpretation accuracy, and fidelity to the source material.

The hard part is not the science; it's the rubric. A task that admits two defensible answers, or a rubric criterion that a grader can read two ways, gets sent back. Feedback arrives asynchronously from project coordinators, and revision cycles are part of the job.

What the screen looks for

  • Real bench or analysis time. Two years of hands-on research with primary data, not just coursework. Expect follow-ups on specific assays, instruments, and how you handled messy data.
  • Diagnostic instinct. Can you look at a methods section and name what's missing — the absent control, the pseudoreplication, the underpowered n?
  • Writing discipline. Scientific prose in clear English, precise enough that a non-specialist grader can apply your rubric without asking you what you meant.
  • Comfort with file formats. CSVs, spreadsheets, PDFs, supplementary documents.

No AI or machine learning background is required or expected.

Logistics

Fully remote contractor engagement, entirely asynchronous — no fixed hours, no standing calls. Compensation is per accepted task rather than hourly; the $30–50/hr band reflects what experts have observed once their workflow settles, and early tasks usually run slower. A weekly minimum submission volume applies. micro1 moves fast on hiring: roles often fill within 48 hours, and they expect first tasks within a day or two of onboarding.