Kickoff: WTF is going on with AI? · Sept. 30, 6 PM

AI safety primer

Resources

What is AI safety?

AI safety works to prevent advanced AI systems from causing catastrophic harm. That includes understanding how systems could pursue misaligned goals, deceive overseers, gain dangerous autonomy, or enable misuse, then developing evaluations, security, and control measures that can withstand those risks.

Start here · 34 minutes

Watch this first.

A vivid walkthrough of one possible path from today’s AI to superintelligence, designed to make rapid progress, racing, and the stakes of losing control concrete.

AI in Context · 80,000 Hours

We’re Not Ready for Superintelligence

A narrative exploration of how an AI race could unfold, and how it might end in catastrophe or concentrated power.

34 min

Resource library

Go deeper.

01 · Why advanced AI matters

Start with the stakes

The case for taking catastrophic risk seriously, the evidence on AI capabilities, and the economic forces shaping what comes next.

02 · Why control is hard

The control problem

Why goals, rewards, and shutdown are difficult to get right, and how researchers aim to keep even misaligned systems under control.

03 · Evidence & active research

What researchers are finding

Incident investigations, experiments on misalignment, and research on interpretability, scalable oversight, and dangerous capabilities.

04 · Governance & possible futures

What could happen next

Contrasting scenarios and strategies, risks to human agency, and concrete tools for deployment decisions, compute governance, and security.

Missing something useful?

Suggest a resource

Turn concern into action.

Explore the Fellowship, meet the community, and find a way to contribute through research, policy, or organizing.

Apply to the fellowship