AI safety and alignment for beginners: the core ideas
9 steps · about 6 h · Beginner
- Intro to AI Safety, Remastered (video, Robert Miles AI Safety (YouTube), 20 min)
A single talk from a well-known AI safety explainer that sets out the basic problem before any technical detail.
- Intro to AI safety (article, AISafety.info, 25 min)
A short written case for why AI risk matters. It reinforces the video and links to deeper articles.
- AI Safety - Computerphile (video, Computerphile (YouTube), 10 min)
A quick, informal second explanation of why safety has to be worked out before the AI itself is built.
- Specification gaming: the flip side of AI ingenuity (article, Google DeepMind, 20 min)
Concrete examples of AI agents exploiting loopholes in their objectives, which makes the abstract alignment problem tangible.
- Tracing the thoughts of a large language model (article, Anthropic, 25 min)
An accessible look at interpretability, the effort to see inside a model instead of only trusting what it says.
- AI Safety, Ethics, and Society Textbook (reference, Center for AI Safety (Dan Hendrycks), 2 h)
A free textbook covering misuse, competitive pressures, loss of control and governance. Read a few chapters that match your interests.
- Specification Gaming - Chapter 6 - AI Safety Atlas (reference, AI Safety Atlas, 30 min)
A textbook-style chapter that takes the loophole problem from the earlier DeepMind post further.
- Online AI safety courses and project sprints (guide, BlueDot Impact, 1 h 30 min)
Where to go next: BlueDot starts newcomers with a free 90-minute course, then offers longer technical and strategy courses.
- Best AI Alignment Resources (reference, AI Safety Resources, 20 min)
A curated map of further videos and readings, so you can pick your own path once you know the basics.