All evaluation work is submitted through the Sovrano AI platform. Here's what to keep in mind:
- Review your submissions carefully before finalising — some task types allow only one submission per item.
- Your work goes through a quality review process. Submissions may be approved, sent back for edits, or flagged.
- Maintain consistent quality throughout your session. Rushing through tasks to increase volume will be noticed and may affect your project standing.
- If you're unsure about a task, use the project's communication channels (typically Discord) to ask your project lead.
Quality expectations
Your justifications and rationales should be specific and evidence-based. "Response A was better overall" will not be accepted. "Response A correctly identifies three risk factors while B only mentions one" will. See our RLHF Guide for detailed examples of good vs. poor justifications.