What goes in
A project starts with a capability or behavior to evaluate. Inputs can include task instructions, representative cases, decision criteria, edge cases, reference materials, and model responses.
Expert data for AI
Expert data for AI captures how qualified people handle real cases, explain decisions, demonstrate tasks, and review model outputs.
Discuss a projectFrom source to use
How it works.
- Contribute
Domain expertise
- Review
Reviewed cases and demonstrations
- Deliver
Training and evaluation data
How the data is made
Experts contribute demonstrations, annotations, rationales, comparisons, or corrections. Review checks the task, domain judgment, and intended use. The process can fit different domains, languages, geographies, and models.
What you receive
The output is a reviewed dataset for the agreed training or evaluation use. Its format, labels, context, and quality records follow the project specification.
Useful distinctions
Questions about expert data.
What kinds of expert data can you create?
Projects can include demonstrations, annotated cases, preference comparisons, decision rationales, model-response reviews, and evaluation sets. The format follows the capability being studied.
How do you choose contributors?
Criteria come from the work: relevant domain experience, language or local context, and the judgment needed for the task. Review requirements follow the same criteria.
Can the work support a specific model or domain?
Yes. Task design, examples, review, and delivery can be built around a model behavior, subject area, language, or operating context.
Have a difficult project?