80,000 Hours
Day 35 after signup
Recent AI progress wasn’t overhyped. Almost everyone underestimated it.
We expected to see dramatic AI progress in the very near future, but so far, it’s been even faster than we thought.
Why it works
- Lead with a timely, specific claim to give the issue a clear editorial focus.
- Place related articles and practical opportunities after the main story so readers can explore the topic or take action.
- Use short section headings to make a long roundup easy to scan.
How it is built
- Editorial headline
- AI progress analysis
- Related reading links
- Featured podcast
Email copy
“Will we have AGI by 2030?” — an update Share: Link Avital Morris October 30, 2026 We expected to see dramatic AI progress in the very near future, but so far, it’s been even faster than we thought. Last year, our colleague Benjamin Todd published “Will we have AGI by 2030?” arguing that there was a significant chance we’d see “a system capable of doing almost all cognitive tasks better than humans” by 2030. Some sceptics thought those predictions sounded like hype — made up by AI companies to sell more AI products and believed by gullible people. This week, Benjamin published an update, and he thinks he was wrong — just not in the direction most sceptics thought. He found that AI progress has actually been faster than almost anyone expected, including him. A few highlights: Superforecasters thought AI would match a top team of virologists on a lab troubleshooting test in 2034. It likely did in April 2025. AI researchers thought AI wouldn’t solve an unsolved maths problem, like a Millennium problem, until about 2054 — but it seems like it happened in September 2026 (although fully validating the solution will take time). Benjamin himself thought there was only a 7% chance it’d happen by the end of 2027. AI is still far from superhuman at physical tasks, but it's progressing faster than most experts thought it would here too. AI researchers thought it would take until 2034 for a bipedal robot to beat the fastest human in a 5 km race, yet in April 2026 a robot beat the human records for the half marathon, as well as the 100 m sprint. Almost everyone is surprised by this progress, even people like us who’ve been predicting rapid progress. Most discussions of how to adapt to AI assume its capabilities are only a little ahead of where they are today. Instead, we need to plan for not only continued rapid progress, but also progress that’s even faster than expected. Read more → The updated “Will we have AGI by 2030?” article Benjamin’s Substack post on how almost everyone underestimated AI For personalised advice on how you can help the world cope with these developments, you can talk to our AI advisor (which is continuing to take advantage of faster-than-expected progress and becoming increasingly useful) or apply to speak to one of our human advisors. Featured release: new video podcast Luisa Rodriguez explains why AI companies won’t slow down on their own, and what could make them. Watch the episode → More new releases: Tom Milton on how we could verify a US–China deal on AI It’s time for frontier AI staff to reassess their work: 8 strategic updates by Robert Wiblin What we’ve been reading: Anthropic files for potentially the biggest IPO ever as CEO warns of risk to humanity from AI by Barbara Ortutay and Bernard Condon What I saw inside Anthropic by Kelsey Piper Red-teaming OpenAI’s internal agent monitoring systems by David Rein and Adam Kaufman The cyberpocalypse is here — sort of by Shakeel Hashim, Veronica Irwin, and Celia Ford A congressional commission on AI risk is overdue by Zoë Baird Opportunities to get involved: Pivotal Research is accepting applications for its Q1 2027 Fellowship. This is a 15-week, full-time research programme in London focused on technical AI safety, technical governance, and AI governance. Fellows work with a mentor and a research manager to produce a research output and plan their next career steps. Apply by November 1. Fourteen AI safety organisations, including 80,000 Hours, are running a joint legal hiring round for full-time and fractional in-house counsel — all in one application! Most roles require a US-barred attorney, and applicants with at least five years of legal practice are more likely to succeed. Apply by November 1. Applications are open for the AI Security Bootcamp, a seven-day intensive programme in San Francisco for security professionals. It covers the threat landscape of frontier AI systems and the hands-on skills to secure them. Apply by Novembe