What the work actually involves

Generalist queues are the catch-all of AI evaluation. On a given day you might be writing adversarial prompts to probe a model's factual reliability, ranking two responses against a rubric and justifying the choice in a short written rationale, rewriting a model answer into the version a careful expert would have given, or labelling where a chain of reasoning first goes wrong rather than simply marking it wrong. Tasks arrive in batches with their own instructions, and those instructions change — a rubric that rewarded thoroughness last week may penalise verbosity this week. The people who last on generalist work are the ones who reread the guidelines each batch instead of running on memory.

Because the label is 'generalist', the projects are not uniform. Some batches skew technical (light coding, spreadsheet logic, data interpretation), some skew writing and editing, some skew everyday practical reasoning where the model's failure is subtle rather than dramatic. Pay tends to sit at the lower end of the band for straightforward comparison work and moves up when a queue requires domain knowledge you can demonstrate.

What the platform screens for

micro1 runs an AI-conducted interview before human review. It probes for: whether you can hold a claim under follow-up questioning, whether your reasoning is written clearly enough to be auditable by someone else, and whether your stated availability matches what you'll actually deliver. Vague, over-broad self-description is the most common reason a generalist application stalls — the screen is looking for at least one area where you go genuinely deep, plus the flexibility to work outside it.

Logistics

  • Fully remote and largely asynchronous; tasks are claimed from a queue rather than scheduled.
  • Part-time is normal — many contributors work 10–20 hrs/week alongside other commitments.
  • Reliable internet and a personal computer are assumed; some projects require screen recording or a specific browser.
  • Work is contract-based per project, with no guarantee of continuous volume between batches.