- Hit
- Missed
- On track
- Pending
The record
every claim by target date · stamp = current verdict · click a row to open its dossierMissed
Radiologists obsolete
Deep learning outperforms radiologists within five years, so we should stop training them now
MissedUS radiologist workforce headcount vs. 201617.3 % growth in US radiologist headcount, 2014-2023 (as of 2023)
Verified blind: DeepSeek V4 + Grok — agree
How it's gradedMet if AI had displaced human radiologists (falling demand / headcount) by ~2021; failed if the workforce kept growing and AI became a complement
US radiologist workforce headcount vs. 20161 measured point 17.3% growth in US radiologist headcount, 2014-2023as of 2023
The other readings measure something else, so they are not joined into a line. All of them are in the table.
Receipts · 3People should stop training radiologists now. It’s just completely obvious within five years deep learning is going to do better than radiologists
UAB Reporter (quoting Geoffrey Hinton, 2016) archived↗context · not model-verified · 2022-12
the number of radiologists increased 17.3% (from 30,723 to 36,024)
a significant radiology workforce shortage
Primary sourceRecording: Creative Destruction Lab archived↗ · 2016. The speaker’s own words are in the recording; the quote text above comes from the press report.
Ledger history
- 2026-06-24: seeded
- 2026-06-25: demoted pending bulletproof re-grade
- 2026-07-02: Target date (2021) long past. Neiman HPI (measurement): radiologist headcount grew 17.3% (30,723 to 36,024) from 2014-2023; Nature (2025) corroborates a persistent shortage. Graded wrong per rubric.
- 2026-08-17: Verdict 'wrong' remains correct: the 2021 target is long past and the radiologist workforce grew rather than collapsed. A fresher Neiman HPI release (Feb 2025, neimanhpi.org) refreshes the workforce…
- 2026-09-14: Verdict 'wrong' remains correct (2021 target long past; the workforce grew rather than collapsed). A fresh read of the Neiman HPI release (fetched 2026-09-14) provides a concrete headcount…
- 2026-11-03: The verdict 'wrong' remains correct (2021 target long past; the workforce grew rather than collapsed). A fresh read of the Neiman HPI release (fetched 2026-11-03) provides the same concrete…
- 2026-11-05: Linked the primary source: the organiser’s recording of the talk (Creative Destruction Lab). The quote text still comes from the press report in the receipts. Verdict unchanged.
Hit
IMO gold medal
An AI built before the contest earns a gold medal at the International Mathematical Olympiad by 2025
HitIMO score vs. the gold cutoff (35/42 in 2025)35/42 points at IMO 2025, gold-medal standard (as of 2025-07-21)
Verified blind: DeepSeek V4 + Grok — agree
How it's gradedMet when, at an IMO through 2025, an AI built beforehand scores at or above the gold-medal cutoff under competition conditions
IMO score vs. the gold cutoff (35/42 in 2025)1 measured point 35/42points at IMO 2025, gold-medal standardas of 2025-07-21
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (points at IMO 2025, gold-medal standard) Receipt 2025-07-21 35/42 source Receipts · 3An advanced version of Gemini Deep Think solved five out of the six IMO problems perfectly, earning 35 total points, and achieving gold-medal level performance.
Google DeepMind archived↗ · 2025-07-21
In July 2025, we reached gold medal-level performance on the International Mathematical Olympiad with a general-purpose reasoning model (35/42 points).
So I think we have Paul at <8%, Eliezer at >16% for AI made before the IMO is able to get a gold (under time controls etc. of grand challenge) in one of 2022-2025.
LessWrong — IMO challenge bet with Eliezer (Paul Christiano) archived↗context · not model-verified · 2022-02-25
Ledger history
- 2026-06-24: seeded
- 2026-06-25: demoted pending bulletproof re-grade
- 2026-06-25: The claim targets 2025. Google DeepMind's official blog (source 0) confirms an advanced Gemini Deep Think scored 35/42 at IMO 2025, meeting the gold-medal cutoff. OpenAI's First Proof blog (source 1)
- 2026-06-26: added the Yudkowsky–Christiano bet-odds source (LessWrong, >16% vs <8%) to the receipt; context only — claim, grade, and the 2-model verification binding are unchanged
- 2026-10-24: Verdict 'ahead'/'met' remains correct (target 2025 passed with gold-standard performance). A fresh read of the DeepMind blog confirms the same 35/42 measurement; refreshing the receipt.
Missed
AI smarter than any human by end of 2025
AI smarter than any one human probably around the end of 2025
MissedBroad cognitive capability vs. the single most capable human0% Pure LLM score on ARC-AGI-2 (as of 2025-03-24)
Verified blind: GLM-5.2 + Grok — agree
How it's gradedMet if, by end of 2025, a deployed AI system demonstrably exceeds the most capable individual human across the breadth of cognitive tasks (not just benchmarks); failed if leading systems remain below top human experts in substantial domains
Broad cognitive capability vs. the single most capable human1 measured point 0%Pure LLM score on ARC-AGI-2as of 2025-03-24
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (Pure LLM score on ARC-AGI-2) Receipt 2025-03-24 0% source Receipts · 3“My guess is that we’ll have AI that is smarter than any one human probably around the end of next year,” Musk said in a livestreamed interview on his social network X.
The Guardian archived↗context · not model-verified · 2024-04-09
best AI agents are not currently able to carry out substantive projects by themselves or directly substitute for human labor.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
Pure LLMs score 0% on ARC-AGI-2
ARC Prize — ARC-AGI-2 and ARC Prize 2025 archived↗ · 2025-03-24
Primary sourceRecording: Norges Bank Investment Management (In Good Company with Nicolai Tangen) · 2024-04. The speaker’s own words are in the recording; the quote text above comes from the press report.
Ledger history
- 2026-07-05: Target date 2025-12 is past. The claim required AI to exceed the most capable human across the breadth of cognitive tasks by end of 2025. Two recognized-domain measurements show this did not occur…
- 2026-11-03: The verdict 'wrong' remains correct (target 2025-12 passed without AI exceeding the most capable human across cognitive domains). A fresh read of the ARC Prize source confirms a concrete measurement…
- 2026-11-05: Linked the primary source: the recording of the April 2024 X Spaces interview, published by Norges Bank Investment Management. The quote text still comes from the press report in the receipts. Verdict…
Missed
AI agents join the workforce in 2025
In 2025, the first AI agents join the workforce and materially change the output of companies
MissedDocumented production deployments of autonomous agents with material output impact6:1 interactive AI agent hours to employee hours at Anthropic (frontier-adopter case study) (as of 2026-08)
Verified blind: GLM-5.2 + Grok — agree
How it's gradedMet if by end of 2025 autonomous AI agents are deployed in real companies doing economically material work (documented output changes), beyond copilots and chat assistants; failed if agent deployments remain pilots without material output impact
Documented production deployments of autonomous agents with material output impact1 measured point 6:1interactive AI agent hours to employee hours at Anthropic (frontier-adopter case study)as of 2026-08
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (interactive AI agent hours to employee hours at Anthropic (frontier-adopter case study)) Receipt 2026-08 6:1 source Receipts · 3We are now confident we know how to build AGI as we have traditionally understood it. We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies.
Sam Altman — Reflections (blog) archived↗context · not model-verified · 2025-01
best AI agents are not currently able to carry out substantive projects by themselves or directly substitute for human labor.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
The 95% failure rate for enterprise AI solutions
Fortune — MIT NANDA report on GenAI pilots archived↗ · 2025-08-18
Ledger history
- 2026-07-15: Target date 2025-12 is past. Two recognized-domain measurements confirm AI agents did not materially change company output: METR (metr.org, measurement) states agents cannot carry out substantive…
- 2026-10-22: Verdict 'wrong' remains correct: the 2025-12 deadline passed without documented broad agent deployments materially changing company output (METR measured in March 2025 that agents could not…
Missed
AI writes 90% of code
Within three to six months AI will be writing 90% of code, and within a year essentially all of it
MissedShare of production code generated by AI vs. written by humans30.1 % of Python functions (US contributors) written by AI, December 2024 (as of 2024-12)
Verified blind: DeepSeek V4 + Grok — agree
How it's gradedMet if, broadly across software, AI is generating ~90% of code by late 2025; failed if human-written code stays the clear majority of production software
Share of production code generated by AI vs. written by humans1 measured point 30.1% of Python functions (US contributors) written by AI, December 2024as of 2024-12
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (% of Python functions (US contributors) written by AI, December 2024) Receipt 2024-12 30.1 source Trajectory unverified indicators — not graded receipts- Sep 2025By late 2025: AI writing ~90% of code (the 3–6 month call) missedSix months on, no verified industry-wide 90% figure materialized; independent analyses put AI's share far lower.assessed Jun 2026indication↗
- Mar 2026By Mar 2026: AI writing essentially all code (the 1-year call) missedA year on, human-written code remains the clear majority of production software; even Anthropic's internal 90% claim is disputed.assessed Jun 2026indication↗
Receipts · 4I think we’ll be there in three to six months—where AI is writing 90 percent of the code. And then in twelve months, we may be in a world where AI is writing essentially all of the code.
Council on Foreign Relations (transcript: CEO Speaker Series with Dario Amodei, 2025-03-10) archived↗context · not model-verified · 2025-03-10
AI could soon write 90% of software code in as few as three to six months, Anthropic CEO Dario Amodei said at a Council on Foreign Relations event on March 10.
Yahoo Finance (Amodei at the Council on Foreign Relations, March 2025) archived↗context · not model-verified · 2025-03
By December 2024, AI wrote an estimated 30.1% of Python functions from U.S. contributors
arXiv (Daniotti et al.) archived↗ · 2025-06
20% to 30% of code inside the company’s repositories was “written by software” — meaning AI
TechCrunch archived↗ · 2025-04-29
Primary sourceTranscript: Council on Foreign Relations archived↗ · 2025-03-10. The prediction quote above is taken from it verbatim.
Ledger history
- 2026-06-24: seeded
- 2026-06-25: demoted pending bulletproof re-grade
- 2026-06-26: added trajectory checkpoints — both 3–6mo and 1-yr milestones missed
- 2026-07-03: The claim's target date (2026-03) is in the past. The prediction was that AI would be writing 90% of code by late 2025/early 2026. Source 7 (arXiv, measurement) shows that by December 2024, AI wrote…
- 2026-09-14: Verdict 'wrong' remains correct (target 2026-03 passed with AI's share of production code far below 90%). A fresh read of the arXiv study supplies the best quantified independent measurement of…
- 2026-11-03: The verdict 'wrong' remains correct (target 2026-03 passed with AI's share of production code far below 90%). A fresh read of the arXiv study supplies the best quantified independent measurement of…
- 2026-11-05: Prediction receipt now quotes CFR’s own transcript of the 2025-03-10 event, verbatim and archived; the Yahoo Finance report is kept as corroboration. Verdict unchanged.
- Today · Nov 2026
On track
Superhuman coder
A superhuman coder exists by end of 2027
On trackSWE-bench Verified %79.20 % resolved on SWE-bench Verified (top agent, bash-only default view) (as of 2026-11-04)
2025-042027-121y 1mo on the clockVerified blind: GLM-5.2 + Grok — agree
How it's gradedMet when a model autonomously completes a non-trivial PR end-to-end at senior-eng level
SWE-bench Verified %4 measured points Values: % resolved on SWE-bench Verified (top agent, bash-only default view). Each point is a dated receipt. Open a point for its archived source. Trajectory unverified indicators — not graded receipts- Jun 2025Mid-2025: stumbling agents — first usable AI coding agents appear hitCoding agents emerged in 2025 and now autonomously resolve real GitHub issues on SWE-bench Verified.assessed Jun 2026indication↗
- Jan 2026Early 2026: coding automation accelerates on paceTop agents now exceed 85% on SWE-bench Verified, up from ~70% a year earlier — fast progress, still short of autonomous senior-eng PRs.assessed Jun 2026indication↗
- Dec 2027End 2027: a superhuman coder exists (the target) pendingThe claim's target milestone — not yet due.indication↗
Receipts · 3Why we forecast a superhuman coder in early 2027
AI 2027 (Kokotajlo, Alexander, Larsen, Lifland & Dean) archived↗context · not model-verified · 2025-04
Claude 4.5 Opus (high reasoning) | 76.80
SWE-bench Verified leaderboard archived↗ · 2026-07-15
best AI agents are not currently able to carry out substantive projects by themselves or directly substitute for human labor.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
Ledger history
- 2026-06-20: seeded (#1)
- 2026-06-24: replaced placeholder evidence with the SWE-bench Verified leaderboard; verdict held on-track
- 2026-06-25: demoted pending bulletproof re-grade
- 2026-06-26: added trajectory checkpoints (AI-2027 milestones, on track) + refreshed SWE-bench measurement to >85%
- 2026-07-15: Target date is 2027-12 (future). SWE-bench Verified shows top agents above 76% (swebench.com, measurement, SWE-bench), while METR (measurement, different org) confirms AI agents still cannot…
- 2026-08-15: Verdict on-track remains correct (target 2027-12 is in the future), but a fresh read of the SWE-bench Verified leaderboard (fetched 2026-08-15) shows the top bash-only entry at 79.20% resolved…
- 2026-09-10: Verdict on-track remains correct (target 2027-12 in the future). A fresh read of the SWE-bench Verified leaderboard (fetched 2026-09-10) confirms the top bash-only entry at 79.20% (Claude 4.5 Opus…
- 2026-10-16: Verdict on-track remains correct (target 2027-12 is in the future). A fresh read of the SWE-bench Verified leaderboard (fetched 2026-10-16) confirms the top bash-only entry is unchanged at 79.20%…
- 2026-11-04: Verdict on-track remains correct (target 2027-12 is in the future). A fresh read of the SWE-bench Verified leaderboard (fetched 2026-11-04) confirms the top bash-only entry is unchanged at 79.20%…
Pending
Weakly-general AI
The first weakly general AI system is publicly announced around 2028
PendingPublic announcement date vs. the live community medianunknown — pending a direct read of the live Metaculus median
202020282y 1mo on the clockHow it's gradedMet if a system meeting the Metaculus weakly-general-AI resolution criteria is publicly announced near the community median
ReceiptsNo verified evidence yet.
Ledger history
- 2026-06-24: seeded; awaiting a verifiable read of the Metaculus median before grading
On track
Turing test passed by 2029
AI reaches human-level intelligence and passes the Turing test by 2029
On trackRigorous adversarial Turing-test pass35/42 IMO 2025 score (gold-medal standard) (as of 2025-07-21)
2024-0620293y 1mo on the clockVerified blind: GLM-5.2 + Grok — agree
How it's gradedMet if before end of 2029 an AI passes a rigorous adversarial Turing test (expert judges, extended sessions) or an equivalent accepted human-level-intelligence demonstration; failed otherwise
Rigorous adversarial Turing-test pass1 measured point 35/42IMO 2025 score (gold-medal standard)as of 2025-07-21
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (IMO 2025 score (gold-medal standard)) Receipt 2025-07-21 35/42 source Receipts · 3solved five out of the six IMO problems perfectly, earning 35 total points
Google DeepMind — Advanced Gemini Deep Think at IMO 2025 archived↗ · 2025-07-21
They outperform experts on most exam-style problems for a fraction of the cost.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
I have stayed consistent. So 2029, both for human-level intelligence and for artificial general intelligence (AGI) – which is a little bit different.
The Guardian / Observer (Kurzweil interview) archived↗ · 2024-06-29
Ledger history
- 2026-08-17: Target date 2029 is in the future, so the verdict is capped at on-track/partial. Two recognized-domain evidence items with different origin_orgs (>=1 measurement) document concrete positive progress…
- 2026-11-03: The verdict 'on-track' remains correct (target 2029 is in the future). A fresh read of the Google DeepMind source confirms the concrete IMO 2025 measurement of progress toward human-level…
On track
AI passes every human test by 2029
Within five years (by early 2029), AI does well on every single test the computer-science industry can put in front of it
On trackCoverage of standardized human tests at strong-pass level35/42 IMO 2025 score (gold-medal standard) (as of 2025-07-21)
2024-032029-032y 4mo on the clockVerified blind: GLM-5.2 + Grok — agree
How it's gradedMet if by March 2029 frontier AI systems achieve strong performance on essentially every standardized human test the field proposes (bar exams, medical boards, olympiads, etc.); failed if significant test categories remain unconquered
Coverage of standardized human tests at strong-pass level1 measured point 35/42IMO 2025 score (gold-medal standard)as of 2025-07-21
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (IMO 2025 score (gold-medal standard)) Receipt 2025-07-21 35/42 source Receipts · 3solved five out of the six IMO problems perfectly, earning 35 total points
Google DeepMind — Gemini Deep Think at IMO 2025 archived↗ · 2025-07-21
They outperform experts on most exam-style problems for a fraction of the cost.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
"If I gave an AI… every single test that you can possibly imagine, you make that list of tests and put it in front of the computer science industry, and I'm guessing in five years time, we'll do well on every single one," Huang said.
Fox Business (Stanford SIEPR Economic Summit remarks) archived↗ · 2024-03-03
Primary sourceRecording: Stanford Institute for Economic Policy Research (SIEPR) archived↗ · 2024-03. The speaker’s own words are in the recording; the quote text above comes from the press report.
Ledger history
- 2026-08-16: Target date 2029-03 is in the future, so on-track/partial is the ceiling. Google DeepMind (deepmind.google, measurement) records an officially graded, gold-medal-standard 35/42 on IMO 2025 — one of…
- 2026-11-03: The verdict 'on-track' remains correct (target 2029-03 is in the future). A fresh read of the Google DeepMind source confirms the concrete IMO 2025 measurement of progress toward passing every test.
- 2026-11-05: Linked the primary source: the organiser’s recording of the talk (Stanford SIEPR). The quote text still comes from the press report in the receipts. Verdict unchanged.
On track
LeCun: human-level AI a decade out
Human-level AI is years away, if not a decade — and will not come from LLMs alone (counterpoint claim)
On trackArrival date and architecture of the first accepted human-level system79.20 % resolved on SWE-bench Verified (top agent, bash-only default view) (as of 2026-11-03)
2024-1020348y 1mo on the clockVerified blind: GLM-5.2 + Grok — agree
How it's gradedMet if human-level AI does NOT appear before ~2030 (vindicating the skeptic call); failed if a widely accepted human-level system arrives well before the end of the decade or emerges primarily from LLM scaling
Arrival date and architecture of the first accepted human-level system1 measured point 79.20% resolved on SWE-bench Verified (top agent, bash-only default view)as of 2026-11-03
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (% resolved on SWE-bench Verified (top agent, bash-only default view)) Receipt 2026-11-03 79.20 source Receipts · 3the best AI agents are not currently able to carry out substantive projects by themselves or directly substitute for human labor.
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
Claude 4.5 Opus | Sonar Foundation Agent | 79.20
SWE-bench Verified leaderboard archived↗ · unknown
“It’s going to take years before we can get everything here to work, if not a decade,” said LeCun. “Mark Zuckerberg keeps asking me how long it’s going to take.”
TechCrunch (Hudson Forum keynote) archived↗ · 2024-10-16
Primary sourceRecording: Hudson Forum archived↗ · 2024-10. The speaker’s own words are in the recording; the quote text above comes from the press report.
Ledger history
- 2026-09-10: Target date 2034 is far in the future, so on-track/partial is the ceiling. LeCun's skeptic call is currently holding: METR (metr.org, measurement, different org) measures that the best agents still…
- 2026-11-03: The verdict 'on-track' remains correct (target 2034 is in the future). A fresh read of the SWE-bench Verified leaderboard (fetched 2026-11-03) confirms the top bash-only entry remains at 79.20%…
- 2026-11-05: Linked the primary source: the organiser’s recording of the talk (Hudson Forum). The quote text still comes from the press report in the receipts. Verdict unchanged.
On track
Hassabis: AGI in 5–10 years
AGI arrives in the next five to ten years (2030–2035)
On trackPublic demonstration of a generally human-level system79.20 % resolved on SWE-bench Verified (top agent, bash-only default view) (as of 2026-11-03)
2025-0420359y 1mo on the clockVerified blind: GLM-5.2 + Grok — agree
How it's gradedMet if a system widely accepted as AGI (general human-level capability across domains) is publicly demonstrated between 2030 and 2035; graded early as wrong only if Hassabis's stated criteria are clearly unmet by end of 2035
Public demonstration of a generally human-level system1 measured point 79.20% resolved on SWE-bench Verified (top agent, bash-only default view)as of 2026-11-03
A single reading is not yet a trajectory. The figure appears once a second measurement lands.
Table view
As of Measured (% resolved on SWE-bench Verified (top agent, bash-only default view)) Receipt 2026-11-03 79.20 source Receipts · 3Claude 4.5 Opus | Sonar Foundation Agent | 79.20
SWE-bench Verified leaderboard archived↗ · 2026-09-03
has been doubling approximately every 7 months for the last 6 years
METR — Measuring AI Ability to Complete Long Tasks archived↗ · 2025-03-19
In the next five to ten years, I think.
CBS News — 60 Minutes interview transcript archived↗ · 2025-04-20
Ledger history
- 2026-09-03: Target date 2035 is well in the future, so on-track/partial is the ceiling. Two recognized-domain evidence items with different origin_orgs (both role:measurement) show concrete capability progress…
- 2026-11-03: The verdict 'on-track' remains correct (target 2035 is in the future). A fresh read of the SWE-bench Verified leaderboard (fetched 2026-11-03) confirms the top bash-only entry remains at 79.20%…