Path
Expert, safety & agents
Check claims of fact. Judge where a subject ends. Report harmful failures. Read the step-by-step trail an AI agent leaves.
- Prepares you for
- It prepares you for safety review and agent-evaluation work. That is the expert end of human-feedback projects.
- What you learn
- You split an answer into claims you can check. You write up a harmful failure without spreading it further. You find the first unsafe step in an AI agent's tool trail. You decide which of an agent's actions need your approval every single time.
- Why this order
- Checking claims is the base skill. Safety reporting, agent traces and policy edges each add one more thing that can go wrong.
- Check claims and evidence24 minutes
- Sort safety failures and send them up25 minutes
- Review an agent trace24 minutes
- Decide what an agent may do alone22 minutes
- Apply a policy at its edge22 minutes
- Ask for a second opinion well20 minutes
- Sort failures by how bad they are22 minutes
- Run a review round with others24 minutes
Curriculum reviewed: