Phase 6 · Modern LLM engineering · Systems & safety
Alignment and safety
Making capable models behave as intended.
Open this lesson Create a free account
Why it matters
As models get more capable and act more autonomously (browsing, running code, sending messages), the gap between what we asked for and what we meant matters more. Alignment research works on closing that gap; safety practice limits the damage when it isn't closed. Both shape the products you use, and both apply directly to anything you build with agents.
What you'll be able to do
- Distinguish misuse, accidents and misalignment as sources of AI risk
- Show specification gaming with a small reinforcement-learning agent, and fix it
- Describe how labs evaluate and manage risk in frontier models
- Apply practical safeguards when building with agents: least privilege, approvals, untrusted input
Before this lesson
Next in this phase: Multi-agent systems