What we send to people who ask how to get into AI safety. Nearly all of it is free, and nearly all of it is in English.
Five pieces to read first, in this order.
The clearest single overview of the risk, and of the kinds of work that address it.
Where the evidence stands on risks from general-purpose AI, written for governments by researchers from across the field.
A concrete scenario of how the next few years could unfold. Useful for making the abstract argument specific.
Why one frontier lab considers the problem urgent, and how it hedges across different outcomes.
The technical case, step by step, for why aligning systems more capable than us is hard.
Free online courses in small groups, from a two-hour introduction to multi-week courses on technical safety and governance.
Part-time, remote research projects with mentors from the field. A common first research experience.
Weekend hackathons, online and at local hubs. The fastest way from reading about a problem to working on it.
A free textbook and course, from catastrophic risks to governance.
Several weeks of hands-on engineering for people who already code: transformers, interpretability, reinforcement learning and evaluations.
A full-time research fellowship for people ready to do alignment research with a mentor.
Patient, precise explanations of the central problems in alignment.
Documentary-length videos on where AI is heading.
Animated adaptations of essays on AI and the long-term future.
Long interviews with alignment researchers about their papers.
In-depth conversations about AI, policy and what to do with a career.
Weekly news on AI policy and safety, for keeping up.
The history of machine learning, told through the difficulty of getting systems to do what we mean.
An argument for rebuilding AI around uncertainty about what humans want.
The pessimistic case at its most direct. Worth reading whether or not it convinces you.
What the community here writes.
Writing from people in the community: event write-ups, reading notes and what is happening in Barcelona and Spain.
Organisations, courses, events, funding and communities across the whole field, kept up to date.
Open roles in AI safety and governance.
Where much of the technical research is posted and argued over first.
The reading group and our events are where these get discussed, and where you can ask about whatever doesn't add up.