What the work actually involves

You write problems that models get wrong, then write the correct solution with an explanation clear enough for a trainer to grade against. In practice that means taking a sales distribution across locations and months and reasoning carefully about which location genuinely grew most — including which time window to use and whether early-period volatility should be discounted. It also means constrained logic puzzles (ordering rules, scheduling, deduction), reading long documents and decomposing them into logical blocks, and verifying claims in text through online research with sources you can point to. A second stream of the work is reviewing model output: identifying where the chain of reasoning breaks, annotating the specific step that fails, and writing feedback the model can learn from rather than a verdict.

Hebrew is not decorative here. Tasks are produced and reviewed across both languages, so you need Hebrew strong enough to write natural, idiomatic prompts and explanations — not translated English — and English comfortable enough to read dense source material and argue about nuance with reviewers.

What the screen looks for

  • Reasoning you can narrate. Screens typically pose a small analytical or logic problem and follow up on how you arrived at the answer. Getting it right silently scores worse than getting it right out loud.
  • Awareness of ambiguity. The sales-growth example in the listing is deliberately underspecified. Candidates who notice the missing assumptions and state them do well.
  • Feedback that localises error. "Wrong" is not annotation. "Correct until it treated the December spike as trend rather than a one-off" is.
  • Genuine Hebrew fluency, probed through follow-up rather than a checkbox.

Logistics

Freelance contractor engagement — no paid leave or medical benefits. Fully remote, 40 hours per week for the contract duration, with 2–5 hours per day overlapping America/Los_Angeles (UTC-8). Pay is undisclosed; Turing describes it as based on experience and expertise, and contract extension depends on performance and project need. You supply your own desktop or laptop and a reliable connection. No prior specialised domain experience is required; professional writing or analyst background is preferred but a degree is not mandatory.