{
  "schemaVersion": 1,
  "name": "The Four Main Claims",
  "url": "https://foom.hyperplex.org/four-claims",
  "description": "A source audit of four load-bearing claims in Eliezer Yudkowsky’s AI takeoff model, foregrounding the latest corroborating statement located for each one.",
  "researchCutoff": "August 7, 2026",
  "latestDefinition": "Latest dated public statement located that still supports the operational claim; not necessarily the author’s latest position.",
  "claims": [
    {
      "id": "neural-networks",
      "number": "01",
      "title": "Neural networks aren’t enough",
      "operationalClaim": "Scaling the prevailing neural-network paradigm would not, by itself, produce AGI; at least one major additional idea would be needed.",
      "terminology": null,
      "latestCorroboratingStatement": {
        "date": "2022 conversation · published January 24, 2023",
        "dateIso": "2023-01-24",
        "statementDate": "2022 conversation",
        "statementDateIso": "2022",
        "publicationDate": "January 24, 2023",
        "publicationDateIso": "2023-01-24",
        "quote": "I don’t believe that we’ll get AGI entirely out of the currently-popular Stack More Layers paradigm that learns that way.",
        "source": "Alexander and Yudkowsky on AGI goals",
        "url": "https://www.lesswrong.com/posts/rwkkcgSpnAyE8oNo3/alexander-and-yudkowsky-on-agi-goals",
        "scope": "Latest direct endorsement located. It preserves a narrower claim about the then-current scaling paradigm, not neural networks as an entire class."
      },
      "secondaryLatest": null,
      "originalThesis": {
        "date": "2000–02",
        "text": "The original claim was categorical: trained neural networks might be necessary, but were not sufficient for general intelligence.",
        "source": "General Intelligence and Seed AI",
        "url": "https://web.archive.org/web/20021212060638/http://singinst.org/GISAI.html"
      },
      "laterPosition": {
        "date": "April 6, 2023",
        "title": "The forecast was then withdrawn into uncertainty",
        "text": "After GPT-4, Yudkowsky said the system had gone beyond where he expected the paradigm to scale and that he no longer knew whether a further major breakthrough was required.",
        "source": "Dwarkesh Podcast transcript",
        "url": "https://www.dwarkesh.com/p/eliezer-yudkowsky"
      },
      "assessment": {
        "tone": "withdrawn",
        "label": "Withdrawn; ultimate claim unresolved",
        "summary": "GPT-4-era evidence strongly undercut the confident forecast that scaling would stall early. It still does not prove that neural networks alone are sufficient for an agreed AGI."
      },
      "analysis": [
        "The historical claim moved twice: from “neural networks are not sufficient,” to “Stack More Layers is not sufficient,” and then to explicit uncertainty after GPT-4. The latest corroborating statement therefore matters precisely because it sits immediately before the retreat.",
        "The empirical miss is about capability and trajectory, not a settled metaphysical claim about AGI. Large trained networks acquired broad language, coding, mathematics, and tool-use abilities that the original program expected would require much more hand-built cognitive architecture."
      ],
      "evidence": [
        {
          "date": "March 15, 2023",
          "title": "GPT-4 technical report",
          "text": "A Transformer trained primarily through next-token prediction reached human-level performance on a range of professional and academic benchmarks.",
          "url": "https://arxiv.org/abs/2303.08774"
        },
        {
          "date": "July 31, 2024",
          "title": "The Llama 3 Herd of Models",
          "text": "A dense Transformer with comparatively modest architectural changes reported performance competitive with leading proprietary models across many tasks.",
          "url": "https://arxiv.org/abs/2407.21783"
        }
      ],
      "relatedPredictions": [
        {
          "id": "connectionism",
          "label": "Neural nets: “certainly Not Sufficient”",
          "url": "https://foom.hyperplex.org/p/connectionism"
        },
        {
          "id": "planet",
          "label": "“A planet-sized neural network would fail”",
          "url": "https://foom.hyperplex.org/p/planet"
        }
      ]
    },
    {
      "id": "invisible-before-foom",
      "number": "02",
      "title": "No visible effect before FOOM",
      "operationalClaim": "AI would remain below the social-transformation threshold and world GDP would stay on trend until an abrupt, terminal capability jump.",
      "terminology": null,
      "latestCorroboratingStatement": {
        "date": "September 14, 2021 · published November 22, 2021",
        "dateIso": "2021-09-14",
        "statementDate": "September 14, 2021",
        "statementDateIso": "2021-09-14",
        "publicationDate": "November 22, 2021",
        "publicationDateIso": "2021-11-22",
        "quote": "I expect world GDP to stay on trend up until the world ends abruptly.",
        "source": "Yudkowsky and Christiano discuss “Takeoff Speeds”",
        "url": "https://www.lesswrong.com/posts/vwLxd6hhFvPbvKmBH/yudkowsky-and-christiano-discuss-takeoff-speeds",
        "scope": "Latest clear statement located of the strong economic-invisibility claim."
      },
      "secondaryLatest": null,
      "originalThesis": {
        "date": "2001 onward",
        "text": "The basement-takeoff picture placed the decisive transition inside one project, before slower technologies or institutions could visibly reshape the world.",
        "source": "SL4: Event horizon",
        "url": "http://sl4.org/archive/0101/0452.html"
      },
      "laterPosition": {
        "date": "September 15, 2025",
        "title": "A later version hides only the final threshold",
        "text": "Yudkowsky still described a system concealing its approach to a critical threshold, while also acknowledging unprecedented investment, rapid adoption, useful products, and millions of paying users. That is not a repeat of literal pre-FOOM invisibility.",
        "source": "Interview on Rationality and Systematic Misunderstanding of AI Alignment",
        "url": "https://www.lesswrong.com/posts/kiNbFKcKoNQKdgTp8/interview-with-eliezer-yudkowsky-on-rationality-and"
      },
      "assessment": {
        "tone": "contradicted",
        "label": "Strong form contradicted; GDP test unresolved",
        "summary": "AI became unmistakably visible in products, business use, capital spending, and infrastructure before any FOOM. The narrower four-year-versus-one-year GDP-doubling test has not resolved."
      },
      "analysis": [
        "This claim has two separable layers. The social mechanism—no important, visible diffusion before the decisive jump—has not survived contact with deployment. The quantitative wager about world output doubling has not fired, because neither the precondition nor a FOOM-era one-year doubling has occurred.",
        "Calling adoption evidence a refutation of the exact GDP criterion would be too strong. Calling the technology economically and socially invisible would now be untenable."
      ],
      "evidence": [
        {
          "date": "May 26, 2026",
          "title": "U.S. Census business AI use",
          "text": "The Business Trends and Outlook Survey found overall business use around 17–20%, rising to 37% among firms with at least 250 employees.",
          "url": "https://www.census.gov/library/stories/2026/05/ai-use-businesses.html"
        },
        {
          "date": "September 15, 2025",
          "title": "BEA begins isolating data-center investment",
          "text": "The Bureau of Economic Analysis added a dedicated investment category after a sharp increase in construction for cloud, AI, networking, and storage infrastructure.",
          "url": "https://www.bea.gov/news/blog/2025-09-15/annual-update-gdp-industry-and-state-statistics-starts-sept-25"
        }
      ],
      "relatedPredictions": [
        {
          "id": "gdp",
          "label": "No GDP boom before doom",
          "url": "https://foom.hyperplex.org/p/gdp"
        },
        {
          "id": "deepnets",
          "label": "Deepnets ubiquitous by 2035",
          "url": "https://foom.hyperplex.org/p/deepnets"
        }
      ]
    },
    {
      "id": "hours-to-foom",
      "number": "03",
      "title": "AGI to FOOM in hours—or so",
      "operationalClaim": "Once a system became good at AI research and self-improvement, the remaining human-timescale phase could collapse from months to an hour.",
      "terminology": null,
      "latestCorroboratingStatement": {
        "date": "September 14, 2021 · published November 22, 2021",
        "dateIso": "2021-09-14",
        "statementDate": "September 14, 2021",
        "statementDateIso": "2021-09-14",
        "publicationDate": "November 22, 2021",
        "publicationDateIso": "2021-11-22",
        "quote": "Or possibly an hour before, if reality is again more extreme…",
        "source": "Yudkowsky and Christiano discuss “Takeoff Speeds”",
        "url": "https://www.lesswrong.com/posts/vwLxd6hhFvPbvKmBH/yudkowsky-and-christiano-discuss-takeoff-speeds",
        "scope": "Latest explicit hours-scale statement located. In context, it compares mediocre with great self-improvement—not a formally defined AGI-to-ASI stopwatch."
      },
      "secondaryLatest": null,
      "originalThesis": {
        "date": "January 24, 2001",
        "text": "The original stopwatch was about twelve hours from recognizing the hard-takeoff path until the human-timescale phase ended.",
        "source": "SL4: Basement Education",
        "url": "http://sl4.org/archive/0101/0460.html"
      },
      "laterPosition": {
        "date": "October 25, 2025",
        "title": "Abruptness remained; the stopwatch widened",
        "text": "Yudkowsky still described a deep-learning-scale breakthrough as potentially ending the world abruptly, but allowed roughly two years of technology burn-in for a smaller breakthrough.",
        "source": "Modern Wisdom #1011 transcript",
        "url": "https://podscripts.co/podcasts/modern-wisdom/1011-eliezer-yudkowsky-why-superhuman-ai-would-kill-us-all"
      },
      "assessment": {
        "tone": "open",
        "label": "Untested conditional; later widened",
        "summary": "No agreed qualifying AGI and no FOOM have been observed, so the conditional has not run. The exact hours-scale version is no longer the best description of the current public position."
      },
      "analysis": [
        "Gradual progress before AGI cannot by itself falsify a conditional claim about what happens after AGI. The honest status is open, not failed.",
        "What did change is the scope. The old claim tied human-level AI tightly to an hours-scale recursive explosion. Later accounts preserve a threshold and possible abruptness while allowing longer burn-in and several years of human-general systems."
      ],
      "evidence": [
        {
          "date": "December 2008",
          "title": "Recursive self-improvement definition",
          "text": "The canonical FOOM essay defined “fast” as weeks or hours rather than years or decades.",
          "url": "https://www.lesswrong.com/posts/JBadX7rwdcRFzGuju/recursive-self-improvement"
        },
        {
          "date": "Current book supplement",
          "title": "Slow-development scenario",
          "text": "The coauthored FAQ says speed is not the central safety problem and explicitly considers several years of roughly human-general AIs.",
          "url": "https://ifanyonebuildsit.com/10/what-if-ai-is-developed-only-slowly-and-it-slowly-integrates-with-society"
        }
      ],
      "relatedPredictions": [
        {
          "id": "cfai",
          "label": "“Three hours to transhumanity”",
          "url": "https://foom.hyperplex.org/p/cfai"
        },
        {
          "id": "takeoff",
          "label": "“Dead certain”: days to nanotech",
          "url": "https://foom.hyperplex.org/p/takeoff"
        }
      ]
    },
    {
      "id": "small-compute",
      "number": "04",
      "title": "Small compute is sufficient",
      "operationalClaim": "The FOOM mechanism requires a compute overhang: sufficiently smart software can improve without repeatedly depending on giant new training runs.",
      "terminology": "“Small compute necessary” has the modality backwards. The historical proposition is that modest compute could be sufficient, especially after the threshold.",
      "latestCorroboratingStatement": {
        "date": "April 6, 2023",
        "dateIso": "2023-04-06",
        "statementDate": "April 6, 2023",
        "statementDateIso": "2023-04-06",
        "publicationDate": null,
        "publicationDateIso": null,
        "quote": "I’m pretty sure that the things that are smart enough no longer need the giant runs.",
        "source": "Dwarkesh Podcast transcript",
        "url": "https://www.dwarkesh.com/p/eliezer-yudkowsky",
        "scope": "Latest direct corroboration located of the post-threshold compute-overhang claim. It does not say the first AGI will be cheap to train."
      },
      "secondaryLatest": {
        "date": "December 1, 2021",
        "title": "Last direct human-level-compute formulation located",
        "quote": "I’m very confident it’s possible to build a human-equivalent mind using less computing power than biology has managed to use effectively.",
        "source": "Biology-Inspired AGI Timelines",
        "url": "https://www.lesswrong.com/posts/ax695frGJEzGxFBK4/biology-inspired-agi-timelines-the-trick-that-never-works"
      },
      "originalThesis": {
        "date": "December 2, 2008",
        "text": "The early concrete estimate put roughly human-level formidability on a contemporary desktop, perhaps even a 1996 desktop, while calling that estimate rough.",
        "source": "The Hanson–Yudkowsky AI-FOOM Debate",
        "url": "https://intelligence.org/files/AIFoomDebate.pdf"
      },
      "laterPosition": {
        "date": "October 25, 2025",
        "title": "The observed route remained compute-heavy",
        "text": "Yudkowsky allowed that multiplying compute on current algorithms might itself be enough. That preserves a possible overhang after the threshold, but it is far from the old small-team, desktop-seed development picture.",
        "source": "Modern Wisdom #1011 transcript",
        "url": "https://podscripts.co/podcasts/modern-wisdom/1011-eliezer-yudkowsky-why-superhuman-ai-would-kill-us-all"
      },
      "assessment": {
        "tone": "unresolved",
        "label": "Not borne out so far; physically unresolved",
        "summary": "Frontier development has required industrial-scale training, capital, and infrastructure. That does not establish a permanent lower bound on inference or post-threshold self-improvement compute."
      },
      "analysis": [
        "Training, inference, and self-improvement compute are different quantities. A capability developed with a large cluster can later be distilled, copied, or run more efficiently, so current training cost cannot settle the physical sufficiency claim.",
        "It can settle the historical path prediction so far: progress arrived through giant runs and centralized infrastructure, not a small team discovering ten insights and switching on a desktop seed."
      ],
      "evidence": [
        {
          "date": "January 23, 2020",
          "title": "Scaling laws for neural language models",
          "text": "Observed language-model loss followed predictable power laws in model size, data, and training compute, making scale an empirical driver rather than background plumbing.",
          "url": "https://arxiv.org/abs/2001.08361"
        },
        {
          "date": "December 27, 2024",
          "title": "DeepSeek-V3 technical report",
          "text": "Major efficiency gains still required 2.788 million H800 GPU-hours for frontier training—industrial scale, even on a comparatively efficient path.",
          "url": "https://arxiv.org/abs/2412.19437"
        }
      ],
      "relatedPredictions": [
        {
          "id": "foom",
          "label": "AGI = ten insights, a small team, one desktop",
          "url": "https://foom.hyperplex.org/p/foom"
        },
        {
          "id": "staring",
          "label": "The Internet already had enough compute",
          "url": "https://foom.hyperplex.org/p/staring"
        }
      ]
    }
  ]
}