AI Safety and Alignment: A Grounded Eight-Hour Introduction
by admin · AI safety and alignment
A mastery-oriented overview of the technical, empirical, and governance problems involved in building beneficial AI systems, with fair treatment of skeptical, moderate-concern, and existential-concern positions.
Free account, then it's yours: your own private copy with a clean slate and its own review schedule.
Sign up free to take it →4
modules
12
lessons
8.5h
estimated work · target 8h
What's inside
36 tasks · 30 flashcards · 42 concepts · 36 recall prompts
Sample recall prompts — this is what studying it feels like
- Define AI safety and alignment in your own words, and explain how they differ.
- Name three dimensions that affect AI risk besides raw model capability.
- Give a fair one-sentence version of the skeptic, moderate-concern, and existential-concern positions.
Syllabus
Framing the Problem
Separate safety from alignment, learn why proxies fail, and understand outer versus inner alignment.
- A Map of the AI Safety Debate reading · flashcard · quiz · 5 cards
- Objectives, Proxies, and Specification Gaming reading · quiz · teach_back
- Outer Alignment, Inner Alignment, and Mesa-Optimization reading · flashcard · essay · 5 cards
Technical Approaches in Current Systems
Study human feedback, constitutional training, scalable oversight, interpretability, and their limits.
- Human Feedback and RLHF reading · flashcard · quiz · 5 cards
- Constitutional AI and Scalable Oversight reading · quiz · teach_back
- Interpretability and Mechanistic Evidence reading · note · quiz
Evaluation, Robustness, and Control
Learn how developers search for risky capabilities, stress-test safeguards, and reason about corrigibility and deployment.
- Evals, Red Teaming, and Capability Thresholds quiz · reading · flashcard · 5 cards
- Robustness, Adversarial Use, and Misuse reading · quiz · teach_back
- Corrigibility, Control, and Deployment Norms reading · flashcard · essay · 5 cards
Governance, Disagreement, and Synthesis
Connect technical uncertainty to governance choices and produce a balanced safety brief.
- Governance Levers: Compute, Audits, and Coordination reading · quiz · teach_back
- Serious Disagreements and Empirical Uncertainty reading · note · quiz
- Capstone: Build a Balanced AI Safety Brief project · reading · flashcard · 5 cards