Resources

Start here

We're Not Ready for Superintelligence is a walkthrough of the AI 2027 scenario, one expert view of where things might be headed.

Courses & programs

Two tracks: a general intro on the left, and the technical research pipeline on the right.

Course

BlueDot — The Future of AI

Finish in an evening with a clearer picture of frontier AI, the risks, and what's being done about them. The standard on-ramp into the field.

2 hours·Free·Self-paced

Start the course →

More on the strategy track

Course

BlueDot — AGI Strategy

25 hours · Cohort · Pay-what-you-want

The serious follow-up to Future of AI: development trajectories, possible outcomes, and what strategic interventions might steer things well.

Fellowship

Astra Fellowship — Strategy & Governance

Constellation · 5 months · Berkeley · Paid

In-person stream for catastrophic-risk policy, AI governance, and field-building projects with senior advisors.

Technical

ARENA — Alignment Research Engineer Accelerator

Hands-on technical curriculum: transformers, RL, mech interp, and evals. The standard pipeline for engineers entering alignment research.

~5 weeks·Cohort or self-study·Code-heavy

Explore the curriculum →

Research fellowships

Fellowship

SPAR

Supervised Program for Alignment Research

Remote, part-time, mentor-led research projects. Lower barrier to entry than MATS — great as a first research experience.

Fellowship

MATS

ML Alignment & Theory Scholars

Selective ~10-week paid research program. You're matched with a senior alignment researcher and ship a real project.

Essential reading

The eleven pieces 80,000 Hours considers the best on-ramp to understanding AI risk and where the field is going.

1. 1
   ReportWilliam MacAskill & Fin Moorhouse

   Preparing for the Intelligence Explosion

   How rapid AI advancement could compress centuries of progress into decades — and why that demands preparation now.

2. 2
   ReportKokotajlo, Alexander, Larsen, Lifland, Dean

   AI 2027

   A concrete, near-term AGI scenario built around AI-automated research, with explicit forecasts.

3. 3
   EssayLeopold Aschenbrenner

   Situational Awareness: The Decade Ahead

   The case that AGI may arrive sooner than widely anticipated, with transformative global consequences.

4. 4
   ArticleEge Erdil (Epoch AI)

   The Case for Multi-Decade AI Timelines

   A counterweight to short-timeline takes: why intelligence-explosion scenarios may be overconfident.

5. 5
   EssayHolden Karnofsky

   The Most Important Century

   Argues that transformative AI could make the coming decades the most pivotal in human history.

6. 6
   ArticleAjeya Cotra & Arvind Narayanan

   Does AI Progress Have a Speed Limit?

   Two researchers with contrasting views debate how fast AI is really moving — and what that means.

7. 7
   ReportSevilla et al. (Epoch AI)

   Can AI Scaling Continue Through 2030?

   Projects continued scaling through 2030, constrained primarily by power and chip manufacturing.

8. 8
   PaperJoe Carlsmith

   Existential Risk from Power-Seeking AI

   The canonical careful argument that sufficiently capable, goal-directed AI could pose an existential threat.

9. 9
   PaperKulveit, Douglas, Ammann, Turan, Krueger, Duvenaud

   Gradual Disempowerment

   How advanced AI could erode human agency slowly through institutional drift, not sudden takeover.

10. 10
    PaperRobert Long, Jeff Sebo, et al.

    Taking AI Welfare Seriously

    The case that future AI systems may be moral patients — and what we owe them if so.

11. 11
    EssayDario Amodei

    Machines of Loving Grace

    Anthropic's CEO sketches a positive vision: what powerful AI could do for the world if it goes well.

Curated by 80,000 Hours — read the original list →