What the work involves

CUA data means computer-use agent data: recorded sequences of a human operating a real desktop or browser — clicks, keystrokes, scrolls, window switches — paired with step-level descriptions of intent. Models learn to use software from these trajectories, so a mislabelled step or a skipped state change teaches the wrong behaviour. As a trainer you sit one layer above collection. You review other people's annotated trajectories, write the feedback that makes the next batch better, and turn recurring confusion into documentation rather than answering the same question forty times in chat.

Day to day that looks like: sampling completed tasks against the project rubric, flagging trajectories where the action label doesn't match what the screen actually did, running calibration exercises so a group of annotators converges on the same call for ambiguous cases, onboarding new annotators onto tooling that changes mid-project, and escalating genuine guideline gaps to QA and the PM instead of inventing a local rule. You will also install and run project tools on your own machine under data-handling constraints, which is why the hardware and OS requirements are stated explicitly.

What the screen looks for

Turing's screening is AI-led and pushes on specifics. Expect follow-ups that ask which annotation tool, what the rubric actually said, how many items you reviewed per day, and what your inter-annotator agreement looked like before and after an intervention. Claims about "improving quality" that survive no numbers and no example tend to collapse under two follow-ups. Prior CUA exposure is listed as preferred at 3–4 years; broader annotation QA and trainer experience with a credible account of computer-use workflows is the realistic route in for most candidates.

Logistics

  • Remote, contractor engagement — no paid or medical leave, task-based.
  • 40 hours per week for a stated 4-week duration, with possible extension.
  • 3–4 hours of daily overlap with US Pacific time is required, not optional.
  • Your own machine with 16GB RAM minimum, stable high-speed internet, and Windows, Linux or macOS.
  • Pay is undisclosed on this listing; Turing generally sets hourly rates per project and per region. Treat any figure you hear secondhand as unconfirmed.