Responsible · Intermediate
AI safety and alignment
Alignment methods, misuse risks, privacy, prompt injection and the human oversight that real deployments need.
Why it matters
As systems gain autonomy, failures cost more. Builders need practical safety habits as well as an understanding of the open research problems.
Key ideas
- Alignment and constitutional methods
- Prompt injection and data exfiltration
- Privacy and data handling
- Bias and fairness
- Human oversight
You will be able to
- Threat-model an LLM app
- Defend against prompt injection
- Write a responsible-use policy
Primary reading
- Constitutional AI: Harmlessness from AI Feedback (opens in a new tab)Bai et al. · 2022
- Concrete Problems in AI Safety (opens in a new tab)Amodei et al. · 2016
Learn it here
Planned course