What the work involves

You are writing and reviewing the material that trains and benchmarks frontier models on data work. In practice that means taking a real dataset — Kaggle, UN data, US government open data — and building a task around it: a question worth asking, Python code that answers it defensibly, and a written rationale that makes the analytical choices legible. You will also grade: comparing two or more model responses to a data question, ranking them against a rubric, and writing the explanation of why one is better. Work spans supervised fine-tuning dataset creation, RLHF preference data, evaluation design, and peer review of other contributors' notebooks and documentation.

The unglamorous half is documentation and consistency. A notebook that runs but doesn't explain why a column was dropped, an outlier kept, or a join chosen is a weak training sample even if the output is correct. Reviewers on these projects spend real time on whether the reasoning is reconstructible by someone who wasn't there.

What the screen looks for

  • Working Python fluency for analysis — pandas, numpy, plotting, the ability to go from raw CSV to a defended conclusion without hand-holding.
  • Analytical judgment, not just execution — can you tell a question the data can answer from one it can't, and say so plainly?
  • Rubric discipline — whether your quality judgments are stable and explainable rather than taste-driven.
  • Written English under scrutiny — rationales are the deliverable; unclear writing sinks otherwise good analysis.

Turing's process is stated as roughly 75 minutes: a 60-minute technical round plus a 15-minute cultural and offer conversation. Expect live or near-live code and follow-up questions on your own choices.

Logistics

  • Fully remote, contractor assignment — no medical or paid leave.
  • Commitment tiers of 20, 30, or 40 hours per week; minimum 4 hours per day.
  • Four hours of daily overlap with PST is required, which is the real constraint for candidates in Europe, Africa, or Asia.
  • Contract stated as one month, with start dates that move quickly — the listing mentions next week. Extensions on this kind of work are common but not promised, and pay is undisclosed on this posting.