What the work actually involves

You design short conversational prompts (typically one to five turns) in Thai that force the model to reach into your real personal context — past Gemini chats, Gmail, Google Search and YouTube activity — and then you judge what comes back. Evaluation is broken into named dimensions: Grounding (is every claim about you actually supported, or is it a flawed inference or hallucination?), Integration (is the personal data woven in naturally, or does the model "overnarrate" by announcing where it got everything?), and overall Helpfulness. You also stack-rank two responses side-by-side and write a rationale that explicitly points to the turn where the problem occurred. There is a data-hygiene step too: you extract and verify Debug Info to confirm which sources the model used, then delete evaluation conversations so they don't pollute your own chat history.

What the screen is looking for

Turing's process starts with a Job Interest Form, then a profile review, then a timed assessment that must be returned within 24 hours. The assessment is where most candidates are decided, and it tests judgment rather than credentials: can you tell a correct personalization from a lucky one, can you distinguish a forced connection from a genuinely useful one, and can you write a rationale a reviewer could audit without rerunning the conversation. Thai reading and writing at a high level is the hard gate — the focus language is Thai, and rationales plus prompt design both live in it. Prior data annotation, AI quality evaluation or content moderation experience is preferred rather than required, and the listed degree fields are broad: policy, law, ethics, linguistics, journalism, computer science or similar analytical work.

Logistics and the personal-account condition

  • Contractor engagement, roughly three months, fully remote
  • At least 4 hours a day, up to 40 hours a week, with 4 hours overlapping PST
  • Staffed as a 24-hour global operation, so your local-timezone availability matters as much as total hours
  • Desktop or laptop with a stable connection
  • Observed rate for this project: $15/hour, as stated by the platform

Be deliberate about the personal-account requirement before applying. The evaluation only works if the model has real data to personalize against, so you will connect your primary Google account and enable personal data sources. If that is not something you are comfortable with, this is not a listing to negotiate around — it is structural to the task.