How to handle AI training jobs and do them well

AI training work includes data annotation, response ranking, prompt writing, red-teaming and model evaluation. Demand for it keeps growing, and the gap between average and excellent contributors is large. This guide covers what we teach in our AI Training Assistance program.

1. Treat the guidelines as the spec

Every project ships with guidelines. Read them twice before starting, and keep them open while you work.

• Note edge cases the guidelines mention explicitly.
• When something isn't covered, flag it instead of guessing. Reviewers value good questions.
• Re-read the guidelines after every update. Versions change mid-project.

2. Quality beats speed

Most platforms track accuracy and agreement with reviewers. A fast contributor with low accuracy is removed sooner than a slower, careful one.

Aim for a steady pace you can maintain at 95%+ accuracy, then speed up gradually.

3. Write rationales people can check

For ranking and evaluation tasks, your rationale matters as much as your choice:

1. State the deciding factor first ("Response B contains a factual error about…").
2. Point to the specific sentence or line.
3. Keep it short and neutral.

4. Use your domain expertise

Code, security, finance and legal tasks pay more and need real expertise. Our engineers and security specialists support contributors who take on technical tasks.

5. Protect yourself

• Never share task content outside the platform; most projects are under NDA.
• Take breaks. Some tasks involve disturbing content, so use the platform's opt-out options.
• Track your hours and payments. Our finance team can help with cross-border invoicing.

How we help

Through the program we offer onboarding sessions, guideline walkthroughs, peer review before submission, and a community channel for questions. Talk to the AI Training team to join.