Reading and Courses
Everything worth your time, nothing that isn’t. I’d rather give you ten things you’ll actually use than a hundred you’ll bookmark and forget. Organized by what you’re trying to do, not by topic.
Don't binge this page
Pick the one or two things that match where you are right now. The fastest way to waste this list is to try to consume all of it. (Career strategy (small but matters))
If you’re orienting (start here)
- 80,000 Hours — AI safety technical research — the single best overview of the technical path. Read it once, early.
- 80,000 Hours — building the field of AI safety — for the Field-building and Comms and ops side.
- BlueDot AI Safety Fundamentals — the free intro courses. Do the short one this week.
If you want to learn to build (technical)
- Andrej Karpathy — “Let’s build GPT” / nanoGPT — build a transformer from scratch by following along. The highest-value weekend you can spend. (Deep learning and transformers)
- ARENA curriculum — public, hands-on, the gold-standard path from “can code” to “can do safety research engineering.” (Research Engineer)
- Jacob Hilton — Deep Learning Curriculum — structured self-study for the deeper end.
- 3Blue1Brown — for intuition on linear algebra, calculus, and neural nets. Watch when the math feels scary. (Math you actually need)
If interpretability is pulling at you
- Learn Mechanistic Interpretability — a structured way in.
- Neel Nanda’s “200 Concrete Open Problems in Mechanistic Interpretability” and his getting-started guides (neelnanda.io) — concrete projects you can actually start. (Interpretability)
- ARENA’s interpretability section (in the curriculum above) — the hands-on entry.
If you’re aiming at policy / governance
- GovAI — opportunities hub and a good window into what governance work actually is.
- Horizon Institute for Public Service — the US-government route.
- BlueDot Governance course (same site as above) — the foundational reading-group course.
If you’re about to apply (read before you do)
- Georg Lange — What actually matters in AI safety applications — written by someone who reviewed hundreds of them. Read it before you submit anything. (Applications (the actual mechanics))
- Anthropic Fellows announcement — doubles as a clear list of what current safety research areas even are.
Staying current
- LessWrong and the EA Forum — where a lot of programs are announced and a lot of the field thinks out loud. Good for spotting the next open cohort. (Read critically — it’s a community, not gospel.)
- ERA’s curated opportunities list — periodically updated roundup of safety & governance programs.
Keep your own running list
When you find something genuinely useful, add it here with one line on why. The best version of this note is the one you’ve made your own.
Back to Home · Skills Map · Programs Directory