[Linkpost] “AI 2027: What Superintelligence Looks Like” by Manuel Allgaier

This is a link post. Linkpost from ai-2027.com (with permission of co-author @kokotajlod). For further discussion, see also: their LessWrong post with +555 karma & 106 comments as of now (April 11th) feature in the New York Times (behind paywall) co-authors Daniel Kokotajlo and Scott Alexander on the Dwarkesh podcast.  This is the most well-researched report on a plausible AGI risk scenario that I (Manuel) am aware of. Consider sharing it, especially with people new to AGI risk and with skeptics, I'd love to read good critiques on this. Curious how plausible the scenario and its timeline seems to you, let me know in the comments! Summary (copied from website) We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution. We wrote a scenario that represents our best guess about what that might look like. It's informed by [...] ---Outline:(01:01) Summary(01:28) Daniel Kokotajlos Intro(03:12) Authors(04:08) Report(04:11) Mid 2025: Stumbling Agents(05:50) Late 2025: The World's Most Expensive AI(11:12) Early 2026: Coding Automation(13:26) Mid 2026: China Wakes Up(16:26) Late 2026: AI Takes Some Jobs(18:13) January 2027: Agent-2 Never Finishes Learning(20:58) February 2027: China Steals Agent-2(23:48) March 2027: Algorithmic Breakthroughs(26:35) April 2027: Alignment for Agent-3(30:04) May 2027: National Security(32:27) June 2027: Self-improving AI(34:09) July 2027: The Cheap Remote Worker(37:04) August 2027: The Geopolitics of Superintelligence(43:12) September 2027: Agent-4, the Superhuman AI Researcher(49:12) FootnotesThe original text contained 98 footnotes which were omitted from this narration. --- First published: April 11th, 2025 Source: https://forum.effectivealtruism.org/posts/8iccNXsAdtpYWAtzu/ai-2027-what-superintelligence-looks-like Linkpost URL:https://ai-2027.com/ --- Narrated by TYPE III AUDIO. ---Images from the article:Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.