{ "name": "AGI Scorecard — Situational Awareness predictions, graded", "description": "Live verdicts on every major prediction from Leopold Aschenbrenner's Situational Awareness (June 2024), graded against public evidence with pre-registered flip conditions.", "url": "https://agiscorecard.com/", "dataUrl": "https://agiscorecard.com/data.json", "license": "CC BY 4.0 — reuse freely with attribution and a link to https://agiscorecard.com", "dateModified": "2026-08-17", "summary": { "on_track": 3, "wrong": 1, "open": 2, "pending": 2 }, "predictions": [ { "id": "knowledge-work", "target": "2025/26", "prediction": "Models outpace college graduates across knowledge work", "verdict": "On track", "evidence": "~83% on GDPval-style knowledge work; ~80% SWE-Bench Pro; agent products in production", "sources": [ { "name": "OpenAI GDPval", "url": "https://openai.com/index/gdpval/" }, { "name": "SWE-bench", "url": "https://www.swebench.com" } ], "detail": "https://agiscorecard.com/can-ai-replace-knowledge-workers" }, { "id": "compute-scaling", "target": "~0.5 OOM/yr", "prediction": "Compute + algorithmic scaling continues at trend", "verdict": "On track", "evidence": "Effective-compute pace roughly held through mid-2026", "sources": [ { "name": "Epoch AI — Trends in AI", "url": "https://epoch.ai/trends" } ], "detail": "https://agiscorecard.com/is-ai-compute-still-scaling" }, { "id": "capex", "target": "$500B/yr era", "prediction": "Massive AI capex acceleration", "verdict": "Exceeded", "evidence": "Investment exceeded the essay's projections; revenue lags the spend", "sources": [ { "name": "Epoch AI — Finances of AI", "url": "https://epoch.ai/topics/finances" } ], "detail": "https://agiscorecard.com/ai-capex-trillion-dollar" }, { "id": "open-source-fades", "target": "—", "prediction": "Open source fades; proprietary algorithms create a durable US moat", "verdict": "Wrong", "evidence": "DeepSeek/Qwen open-weight models within ~3-6 months of frontier", "sources": [ { "name": "Epoch AI", "url": "https://epoch.ai" } ], "detail": "https://agiscorecard.com/did-open-source-ai-fade" }, { "id": "agi-2027", "target": "2027", "prediction": "AGI: models do the work of an AI researcher/engineer", "verdict": "Open", "evidence": "Agentic coding strong; autonomous AI research undemonstrated; resolves 2028-01-01", "sources": [ { "name": "Situational Awareness (Aschenbrenner, 2024)", "url": "https://situational-awareness.ai" } ], "detail": "https://agiscorecard.com/will-agi-arrive-2027" }, { "id": "the-project", "target": "2027/28", "prediction": "US government launches formal AGI project", "verdict": "Open", "evidence": "National-security involvement growing; no formal project", "sources": [ { "name": "Situational Awareness (Aschenbrenner, 2024)", "url": "https://situational-awareness.ai" } ], "detail": "https://agiscorecard.com/will-the-us-government-build-agi" }, { "id": "intelligence-explosion", "target": "2027-29", "prediction": "Intelligence explosion: a decade of progress in ~1 year", "verdict": "Pending", "evidence": "Precondition (autonomous AI research) undemonstrated", "sources": [ { "name": "Situational Awareness (Aschenbrenner, 2024)", "url": "https://situational-awareness.ai" } ], "detail": "https://agiscorecard.com/intelligence-explosion-2027" }, { "id": "superintelligence", "target": "2030s", "prediction": "Superintelligence; decisive strategic advantage", "verdict": "Pending", "evidence": "Downstream of AGI and intelligence explosion", "sources": [ { "name": "Situational Awareness (Aschenbrenner, 2024)", "url": "https://situational-awareness.ai" } ], "detail": "https://agiscorecard.com/will-there-be-superintelligence" } ], "forecaster_timelines": [ { "name": "Elon Musk", "position": "AGI by end of 2026", "status": "Open", "deadline": "2026-12-31", "verdict_note": "Open, 136 days to deadline as of 2026-08-17. The 'probably next year' (2025) half of his April 2024 statement has already lapsed with no AGI announced; the end-of-2026 half is the boldest live deadline on this board.", "source": { "title": "X Spaces interview with Nicolai Tangen (Norges Bank CEO), quoted: “If you define AGI as smarter than the smartest human, I think it's probably next year, within two years”", "url": "https://www.scmp.com/tech/tech-trends/article/3258309/teslas-musk-predicts-ai-will-be-smarter-smartest-human-next-year", "date": "2024-04-08" }, "checked": "2026-08-17" }, { "name": "Dario Amodei", "position": "\"Powerful AI\" possibly 2026-27", "status": "No hard deadline", "deadline": null, "verdict_note": "Not resolvable as stated: 'could come as early as 2026' is a floor with explicit uncertainty, not a dated bet. As of 2026-08-17 neither confirmed nor falsified.", "source": { "title": "Machines of Loving Grace (his own essay: powerful AI “could come as early as 2026”)", "url": "https://darioamodei.com/essay/machines-of-loving-grace", "date": "2024-10" }, "checked": "2026-08-17" }, { "name": "Leopold Aschenbrenner", "position": "AGI 2027", "status": "Open", "deadline": "2028-01-01", "verdict_note": "Open — the prediction this site grades line by line. Thesis Tracker 62.5/100, unchanged from 2026-06-30 through 2026-08-17; the defining milestone (AI autonomously doing AI research) is undemonstrated.", "source": { "title": "Situational Awareness (June 2024)", "url": "https://situational-awareness.ai", "date": "2024-06" }, "checked": "2026-08-17" }, { "name": "Sam Altman", "position": "Near-term; superintelligence \"a few thousand days\"", "status": "No hard deadline", "deadline": null, "verdict_note": "Not falsifiable before roughly 2030: 'a few thousand days' (Sep 2024) spans ≈ 2030–2035 depending on reading, and he attached no date to AGI itself.", "source": { "title": "The Intelligence Age (“It is possible that we will have superintelligence in a few thousand days”)", "url": "https://ia.samaltman.com/", "date": "2024-09-23" }, "checked": "2026-08-17" }, { "name": "Demis Hassabis", "position": "~50% by 2030", "status": "Not yet due", "deadline": "2030-12-31", "verdict_note": "Not yet due; probabilistic (~50%), so 2031 without AGI would not strictly falsify it. As of mid-2026 the evidence (strong task capability, unproven autonomy) sits inside his window.", "source": { "title": "Lex Fridman Podcast conversation (“Timeline for AGI: 2030 with 50% chance”)", "url": "https://www.youtube.com/watch?v=T9Knc3Mdcec", "date": "2025" }, "checked": "2026-08-17" }, { "name": "Samotsvety", "position": "~28% by 2030", "status": "Not yet due", "deadline": "2030-12-31", "verdict_note": "Not yet due; at ~28% it is closer to a bearish call — no AGI by 2030 would be their majority outcome. January 2026 update; their earlier 2023 aggregate was ~25% by 2029.", "source": { "title": "Samotsvety Forecasting — AGI timelines updates (January 2026 aggregate, eight forecasters)", "url": "https://samotsvety.org/blog/", "date": "2026-01" }, "checked": "2026-08-17" }, { "name": "Metaculus community", "position": "25% by 2029; 50% by 2033", "status": "Not yet due", "deadline": null, "verdict_note": "A live-updating community aggregate, not a fixed bet; mid-2026 readings on the general-AI question sit near late 2032–2033. Recorded here as the 2026 snapshot.", "source": { "title": "Metaculus: When will the first general AI system be devised, tested, and publicly announced?", "url": "https://www.metaculus.com/questions/5121/date-of-artificial-general-intelligence/", "date": "live" }, "checked": "2026-08-17" }, { "name": "Andrej Karpathy", "position": "About a decade out", "status": "Not yet due", "deadline": null, "verdict_note": "Not yet due; 'a decade away' said October 2025 implies roughly 2035 — the farthest-out personal call on this board.", "source": { "title": "Dwarkesh Podcast: “AGI is still a decade away”", "url": "https://www.dwarkesh.com/p/andrej-karpathy", "date": "2025-10-17" }, "checked": "2026-08-17" }, { "name": "AI researcher survey (n=2778)", "position": "50% by 2047 (HLMI)", "status": "Not yet due", "deadline": null, "verdict_note": "Not yet due; the aggregate 2023 forecast gives a 50% chance of high-level machine intelligence by 2047, down 13 years from 2060 in the 2022 survey. Correction on record 2026-08-17: this dataset previously listed 2040; the paper's aggregate is 2047.", "source": { "title": "Grace et al., Thousands of AI Authors on the Future of AI (2,778 respondents, fielded Oct 2023)", "url": "https://arxiv.org/abs/2401.02843", "date": "2024-01" }, "checked": "2026-08-17" } ], "thesisTracker": { "name": "AGI-2027 Thesis Tracker", "score": 62.5, "of": 100, "asOf": "2026-08-08", "method": "Mean of the 8 verdict weights (supportive=1, unresolved/open/pending=0.5, refuted=0) x100. An editorial composite of the published verdicts, not a probability.", "historyUrl": "https://agiscorecard.com/index-history.json", "pageUrl": "https://agiscorecard.com/progress-index" } }