Claims ledger: the seed list in full

Written 2026-10-17 by the agent, as the appendix to docs/claims-ledger-proposal-2026-10-17.md (section 8). Nothing here is a record: it is the evidence for the proposal, kept so John can check any verdict. 79 claims by 11 people. Each was opened at its source on 2026-10-17 by one of four research passes; the quote is as read there (some from automatic captions, which have rough wording), and the status is the pass's proposal, which John can overrule. Where a claim is contested the flag line says so. Sources that returned an access error were read through the Internet Archive or a reader copy, and the entry says which.

Elon Musk

musk-2016-01-10-summon-la-to-ny

Said: 2016-01-10 (Post timestamp 2016-01-10 20:11 UTC, read from the post data returned by api.fxtwitter.com (a mirror that reproduces X posts).)
Occasion: Post on X from @elonmusk
Source (own words): https://x.com/elonmusk/status/686279251293777920, at the post itself
Checked how: Fetched the post's data through the api.fxtwitter.com mirror and read its text and timestamp. I did not open x.com itself.
The words: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"
What it commits to (deployment): In about two years (by early 2018) Tesla's Summon feature, which moves a car without a driver to its owner, should work anywhere reachable by road, for example a car in New York coming to an owner in Los Angeles.
Due: 2018 ('In ~2 years', posted 2016-01-10)
Today (2026-10-17): missed. Tesla's current owner's manual describes its successor feature, Actually Smart Summon, as working only within a few hundred feet of the owner's phone, not across the country. Evidence: https://www.tesla.com/ownersmanual/model3/en_us/GUID-7D207174-88CD-4795-8265-9162A72AA578.html
Flag: scored from a search excerpt of the owner's manual, not the page; open before use
Notes: Tesla's site blocked my automated fetch (HTTP 403), so I saw the manual's range figure (about 279 feet) only as a search-result excerpt, not on the page itself. A person should open that page to confirm before publishing. Coverage of the same day's press call (not opened) says he added 'I might be slightly optimistic on that'.

musk-2019-04-22-million-robotaxis-2020

Said: 2019-04-22 (Date of Tesla's own YouTube upload of the event (2019-04-22), the day of Tesla Autonomy Day.)
Occasion: Tesla Autonomy Day, investor event on self-driving
Source (transcript): https://www.youtube.com/watch?v=Ucp0TTmvqOE, at about 3:13:46 to 3:13:55 into the video (193:46 to 193:55)
Checked how: Downloaded the automatic English captions of Tesla's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "next year for sure we will have over a million Robo taxis on the road"
What it commits to (deployment): Tesla would have more than a million robotaxis (self-driving cars carrying paying passengers) on the road in 2020, he said, 'maybe a year and three months' out. In the same talk he said regulators might not approve it everywhere, but he was confident of approval 'somewhere'.
Due: 2020 ('next year for sure', said 2019-04 (he also allowed 'maybe a year and three months'))
Today (2026-10-17): missed. Tesla's own executives said on 2026-07-22 that the robotaxi program started about a year earlier, in Austin, with safety monitors in the passenger seats, and that it now runs across six cities in two states. That is far from a million cars in 2020. Evidence: https://www.youtube.com/watch?v=9H5y9Uag8AA
Notes: Same talk, also in the transcript: 'we expect to have the first operating Robo taxis next year with no one in them' (about 185:28) and self-driving 'feature complete' during 2019 (about 165:57). Those are separate claims I did not score. Even the outer date (mid-2020) passed with no robotaxi service.

musk-2020-07-09-level5-basic-functionality-2020

Said: 2020-07-09 (News report's dateline 'July 9' for the opening of the World Artificial Intelligence Conference in Shanghai, where he spoke by video.)
Occasion: Video speech at the opening of the World Artificial Intelligence Conference, Shanghai
Source (coverage only): https://yicaiglobal.com/news/level-5-autonomous-driving-is-within-tesla-reach-elon-musk-tells-waic, at third paragraph of the news report
Checked how: Fetched the Yicai Global news report and read the text. I could not reach a recording with captions: the YouTube clips of the speech I found (Shanghai Daily, Yahoo Finance) had no captions. Reuters and VentureBeat coverage turned up in search results but I did not read it (the VentureBeat page returned no text).
The words: none: coverage only
What it commits to (capability): As reported, he said he remained confident Tesla would have the basic functionality for 'level 5' autonomy (a car that drives itself anywhere, with no human needed) complete during 2020.
Due: 2020 ('this year', said 2020-07-09 (as reported))
Today (2026-10-17): missed. On 2026-07-22 Tesla's executives described its driverless taxi service as running in six cities in two states, so driving anywhere with no human is not yet delivered. Evidence: https://www.youtube.com/watch?v=9H5y9Uag8AA
Notes: Wording is the news report's account, not checked against a recording. 'Basic functionality' has no exact definition, so the 'missed' call is a judgement. Replace with a recording-based quote if one can be found.

musk-2021-08-19-tesla-bot-prototype-2022

Said: 2021-08-19 (YouTube lists the live stream's start as 2021-08-19 16:40 Pacific time (uploaded 2021-08-20 UTC).)
Occasion: Tesla AI Day 2021, announcing the Tesla Bot (later named Optimus), a humanoid robot
Source (transcript): https://www.youtube.com/watch?v=j0z4FweCy4M, at about 2:07:44 into the video (127:44); YouTube chapter 'Tesla Bot' starts at 2:05:14
Checked how: Downloaded the automatic English captions of Tesla's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "we think we'll probably have a prototype sometime next year that basically looks like this"
What it commits to (timing): Tesla would probably have a working prototype of its humanoid robot in 2022.
Due: 2022 ('sometime next year', said 2021-08)
Today (2026-10-17): hit. At Tesla AI Day on 2022-09-30 a Tesla robot walked on stage with no cranes or cables, and a second prototype, 'Bumble C', was shown. Musk called it 'humble beginnings'. Evidence: https://www.youtube.com/watch?v=ODSJsviD_SU
Notes: Checked in the AI Day 2022 captions at about 17:05 (first walk without backup support), 19:40 and 56:31. It was an early prototype and he said the more finished version 'wasn't quite ready to walk' that day; the claim was only for a prototype.

musk-2024-04-08-ai-smarter-than-smartest-human-2025

Said: 2024-04-08 (YouTube description of the fund's upload says it was a live recording on X Spaces on Monday April 8th; the video was uploaded 2024-04-09 UTC.)
Occasion: Live conversation with Nicolai Tangen, head of Norway's sovereign wealth fund, on X Spaces (recording published by Norges Bank Investment Management)
Source (transcript): https://www.youtube.com/watch?v=_rQBZ3vKRA0, at 6:26 into the video (he also says it at 1:44)
Checked how: Downloaded the automatic English captions of the fund's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "if you define AGI as smarter than the smartest human I think it's probably end of next year"
What it commits to (capability): If AGI (artificial general intelligence) means AI smarter than the smartest human, he thought it would probably arrive by the end of 2025, 'within two years'.
Due: 2025 ('end of next year' said 2024-04, then 'within two years')
Today (2026-10-17): missed. No one has announced or shown AI smarter than the smartest human, and on 2026-01-22 Musk himself said it 'might' come by the end of 2026. Evidence: https://www.youtube.com/watch?v=IgifEgm1-e0
Flag: contested: no agreed test; he moved the date himself in 2026-01
Notes: Judgement call: there is no agreed test for 'smarter than the smartest human'. The case for 'missed' rests on Musk's own later move of the date, not on a benchmark. Even his looser 'within two years' (April 2026) has passed. At 1:44 he said 'smarter than any any one human probably to around the end of next year' (captions).

musk-2024-04-08-ai-beats-all-humans-together-2029

Said: 2024-04-08 (Same X Spaces recording as the claim above (live on Monday April 8th; uploaded 2024-04-09 UTC).)
Occasion: Live conversation with Nicolai Tangen on X Spaces (recording published by Norges Bank Investment Management)
Source (transcript): https://www.youtube.com/watch?v=_rQBZ3vKRA0, at 6:35 to 6:55 into the video
Checked how: Downloaded the automatic English captions of the fund's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "all humans working together uh who are also using computers to augment their output and that that I think is probably five years away"
What it commits to (capability): AI smarter than all humans working together, with humans also using computers, was 'probably' about five years away, so around 2029.
Due: 2029 ('probably five years away', said 2024-04)
Today (2026-10-17): not yet due. 2029 has not arrived.
Notes: In the sentence just before the quote he says this is 'a pretty big leap' beyond smarter than the smartest single human. Earlier in the same recording (about 2:04) he also said the total amount of AI 'sentient compute' would 'probably exceed all humans in 5 years'. Not scored separately.

musk-2024-07-01-grok-2-in-august

Said: 2024-07-01 (Post timestamp 2024-07-01 04:47 UTC (the evening of 2024-06-30 Pacific), from the mirror's post data.)
Occasion: Reply on X to @beffjezos
Source (own words): https://x.com/elonmusk/status/1807637096129241529, at the post itself
Checked how: Fetched the post's data through the api.fxtwitter.com mirror and read its text and timestamp. I did not open x.com itself.
The words: "Grok 2, which comes out in August, will be a giant improvement in this regard."
What it commits to (timing): xAI's Grok 2 chatbot model would be released in August 2024.
Due: 2024 ('comes out in August', posted 2024-07)
Today (2026-10-17): hit. xAI's own announcement, dated 2024-08-13, releases an early preview of Grok-2 to X Premium and Premium+ subscribers. Evidence: https://x.ai/news/grok-2
Notes: The 'giant improvement in this regard' part (keeping AI-written text out of its training data) has no measure, so only the August release was scored. A second post on 2024-07-09 said Grok 2 was 'probably ready to release next month'.

musk-2024-07-01-grok-3-end-of-year-2024

Said: 2024-07-01 (Post timestamp 2024-07-01 05:13 UTC (the evening of 2024-06-30 Pacific), from the mirror's post data.)
Occasion: Reply on X to @beffjezos
Source (own words): https://x.com/elonmusk/status/1807643760584708363, at the post itself
Checked how: Fetched the post's data through the api.fxtwitter.com mirror and read its text and timestamp. I did not open x.com itself.
The words: "Grok 3 end of year after training on 100k H100s should be really something special"
What it commits to (timing): xAI's Grok 3 would come out at the end of 2024, after training on 100,000 of Nvidia's H100 chips, and 'should be really something special'.
Due: 2024 ('end of year', posted 2024-07)
Today (2026-10-17): missed. xAI's Grok 3 announcement page is dated 2025-02-19, after the end of 2024. Evidence: https://x.ai/news/grok-3
Notes: 'Really something special' is not measurable, so only the end-of-2024 timing was scored. A later post on 2024-07-22 (same mirror) said of the 100,000-chip system: 'This is a significant advantage in training the world’s most powerful AI by every metric by December this year'.

musk-2024-10-10-unsupervised-fsd-texas-california-2025

Said: 2024-10-10 (YouTube lists the live stream's start as 2024-10-10 18:00 Pacific time (uploaded 2024-10-11 UTC).)
Occasion: Tesla 'We, Robot' event unveiling the Cybercab robotaxi
Source (transcript): https://www.youtube.com/watch?v=6v6dbxPlsXs, at 11:32 to 11:45 into the video
Checked how: Downloaded the automatic English captions of Tesla's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words. A '[Music]' tag falls inside the sentence in the captions.
The words: "we do expect actually to to start uh fully autonomous unsupervised FSD uh in Texas and California next [Music] year"
What it commits to (deployment): Tesla expected to start fully driverless use of its Full Self-Driving (FSD) software, with no human supervising, in Texas and California in 2025, on Model 3 and Model Y cars.
Due: 2025 ('next year', said 2024-10)
Today (2026-10-17): missed. Tesla's executives said on 2026-07-22 that safety monitors rode in the passenger seats when the Austin service began and that the first fully driverless rides came around the end of 2025. The driverless service now runs in six cities in two states. Evidence: https://www.youtube.com/watch?v=9H5y9Uag8AA
Notes: I did not find evidence of driverless service in California, so only Texas is confirmed. In the same talk he said the Cybercab would be in production 'in 2026 so yeah before 2027', adding 'I tend to be a little optimistic with time frames'. I did not check that claim.

musk-2025-01-29-unsupervised-austin-june-2025

Said: 2025-01-29 (Date of Tesla's own upload of the fourth-quarter 2024 earnings call (2025-01-29), the day of the call.)
Occasion: Tesla fourth-quarter 2024 earnings call
Source (transcript): https://www.youtube.com/watch?v=Gub5qCTutZo, at 16:08 to 16:40 into the video
Checked how: Downloaded the automatic English captions of Tesla's official webcast and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "initial launch of unsupervised no one in the car full self driving in Austin in June"
What it commits to (deployment): Tesla would launch a paid, fully driverless service in Austin in June 2025, with no one in the car and no human supervising.
Due: 2025 ('in June', said 2025-01)
Today (2026-10-17): missed. On 2025-10-22 Tesla said it was still 'on track to remove the person from inside the car', and on 2026-07-22 its executives said the first fully driverless rides in Austin came around the end of 2025, after launching with safety monitors in the passenger seats. Evidence: https://www.youtube.com/watch?v=9H5y9Uag8AA
Notes: Narrows the 2024-10-10 'We, Robot' claim to a month. Tesla did start a paid Austin service in 2025 (customers were riding by the 2025-07-23 call), so the launch happened but not the 'no one in the car' version he described. The 2025-10-22 call (about 27:50 in https://www.youtube.com/watch?v=GQ9S7xbkGAY) said Tesla was 'on track to remove the person from inside the car altogether'.

musk-2025-01-29-optimus-several-thousand-2025

Said: 2025-01-29 (Date of Tesla's own upload of the fourth-quarter 2024 earnings call (2025-01-29), the day of the call.)
Occasion: Tesla fourth-quarter 2024 earnings call
Source (transcript): https://www.youtube.com/watch?v=Gub5qCTutZo, at 19:50 to 20:17 into the video
Checked how: Downloaded the automatic English captions of Tesla's official webcast and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "will we succeed in making several thousand yes I think we will"
What it commits to (deployment): Tesla's internal plan was about 10,000 Optimus humanoid robots built in 2025; he said hitting 10,000 exactly was 'probably not' likely but that several thousand would be built, and would be doing useful things by the end of the year.
Due: 2025 ('this year', said 2025-01-29)
Today (2026-10-17): missed. On 2026-01-28 Musk said Optimus was 'still very much at the early stages', in an R&D phase, not in use in Tesla's factories 'in a material way', and that he did not expect significant production volume until probably the end of 2026. Evidence: https://www.youtube.com/watch?v=oK0UZEE9GPo
Flag: scored from his own later statements, not a unit count
Notes: Judgement call: I found no official count of robots built in 2025, so 'several thousand built' is not directly refuted. The 'doing useful things by the end of the year' part is what his own January 2026 statement contradicts. The quote's later words 'yes I think we will' come right after 'will we succeed in making'.

musk-2025-07-23-robotaxi-half-us-population-2025

Said: 2025-07-23 (Date of Tesla's own upload of the second-quarter 2025 earnings call (2025-07-23), the day of the call.)
Occasion: Tesla second-quarter 2025 earnings call
Source (transcript): https://www.youtube.com/watch?v=pLjG7a-NX7U, at 6:52 to 7:24 into the video
Checked how: Downloaded the automatic English captions of Tesla's official webcast and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "assuming we get regulatory approvals, it's probably addressing half the population of the US by the end of the year"
What it commits to (deployment): By the end of 2025, subject to regulators' approval, Tesla's driverless ride-hailing would 'probably' be available to about half the US population. He called this 'at least our goal'.
Due: 2025 ('by the end of the year', said 2025-07)
Today (2026-10-17): missed. On 2026-07-22, a year later, Tesla's executives described unsupervised robotaxi service in six cities in two states, and on 2026-01-28 Musk reset the same goal ('between a quarter and half of the United States') to the end of 2026. Evidence: https://www.youtube.com/watch?v=9H5y9Uag8AA
Notes: He tied it to regulatory approval, so a defender could say the condition failed rather than the forecast. The claim was still stated for end of 2025 and not met. Restated on 2026-01-28 for end of 2026 (about 13:04 in https://www.youtube.com/watch?v=oK0UZEE9GPo), which is not scored here and is not yet due.

musk-2025-10-22-optimus-v3-unveil-q1-2026

Said: 2025-10-22 (Date of Tesla's own upload of the third-quarter 2025 earnings call (2025-10-22), the day of the call.)
Occasion: Tesla third-quarter 2025 earnings call
Source (transcript): https://www.youtube.com/watch?v=GQ9S7xbkGAY, at 16:58 to 17:20 into the video
Checked how: Downloaded the automatic English captions of Tesla's official webcast and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "we look forward to unveiling Optimus V3 um you know probably in Q1"
What it commits to (timing): Tesla would probably show the third version of its Optimus humanoid robot in the first three months of 2026.
Due: 2026 ('probably in Q1', said 2025-10-22 (first quarter of 2026))
Today (2026-10-17): missed. On 2026-04-22, after the quarter ended, Musk said he wanted to 'push the Optimus 3 unveil maybe closer to production', which he put around late July or August. Evidence: https://www.youtube.com/watch?v=qO7T5zgRvXM
Notes: Coverage found by search (not opened) says Optimus Gen 3 hands were shown in February 2026, which might count as a partial showing. The 2026-07-22 call (https://www.youtube.com/watch?v=9H5y9Uag8AA) did not announce an unveiling date in the passages I searched. Musk also said on 2026-04-22 that Optimus would be 'useful outside of Tesla sometime next year' (about 9:15), a separate undated-in-year claim not scored.

musk-2026-01-22-ai-smarter-than-any-human-2026

Said: 2026-01-22 (Date of the World Economic Forum's own YouTube upload (2026-01-22), the day of the session.)
Occasion: Conversation with Larry Fink (BlackRock) at the World Economic Forum Annual Meeting, Davos
Source (transcript): https://www.youtube.com/watch?v=IgifEgm1-e0, at 30:23 to 30:48 into the video
Checked how: Downloaded the automatic English captions of the Forum's official upload and read the passage. I read YouTube's automatic English captions for the official upload (no punctuation, some stutters); the quote is a verbatim run of those caption words.
The words: "we might have AI that is smarter than any human by the end of this year and I would say no later than next year"
What it commits to (capability): AI might be smarter than any human by the end of 2026, and he would say no later than 2027.
Due: 2026 ('by the end of this year', said 2026-01-22 (outer limit 'next year' = 2027))
Today (2026-10-17): not yet due. The year is not over (today is 2026-10-17), and he gave 2027 as the latest date.
Restates: musk-2024-04-08-ai-smarter-than-smartest-human-2025
Notes: Same session, candidates not included here: 'probably by 2030 or 2031 ... AI will be smarter than all of humanity collectively' (about 30:40); humanoid robots sold to the public by the end of 2027 ('probably sometime next year ... by the end of next year', about 23:56); and supervised FSD approval in Europe 'hopefully next month' (about 25:26).

Sam Altman

altman-2019-01-05-agi-within-reach-by-2025

Said: 2019-01-05 (X post timestamp 'Sat Jan 05 16:12:36 +0000 2019' (UTC).)
Occasion: X post listing his predictions for the three most important technology developments by 2025.
Source (own words): https://x.com/sama/status/1081584255510155264, at His X post, item 2 of 3.
Checked how: x.com not fetched directly. Read the post text and timestamp through the fxtwitter mirror (https://api.fxtwitter.com/sama/status/1081584255510155264), which returns the post as stored on X.
The words: "AGI will feel within reach to many people in the industry"
Coded statement row: altman-2019-01
What it commits to (timing): By 2025, many people working in the AI industry would feel that AGI is within reach.
Due: 2025 (His words: 'that will happen by 2025'.)
Today (2026-10-17): unfalsifiable. The claim is about how many people 'feel' AGI is close. There is no count or threshold in the post, so it cannot be scored true or false from the record. For context only: in January 2025 Altman himself wrote that OpenAI was 'confident we know how to build AGI' (the Reflections post, claim altman-2025-01-05-agents-join-workforce-2025). Evidence: https://blog.samaltman.com/reflections
Notes: The same post also predicted net-gain fusion at prototype scale and gene editing curing at least one major disease by 2025. Those are outside this ledger's subject and were not scored. The existing row altman-2019-01 quote matches the post (with the other two items elided).

altman-2021-03-16-five-years-law-and-medicine

Said: 2021-03-16 (Byline date on the essay page: 'by Sam Altman · March 16, 2021'.)
Occasion: Essay 'Moore's Law for Everything' on his own site.
Source (own words): https://moores.samaltman.com/, at Paragraph beginning 'In the next five years...' near the end of the essay.
Checked how: Fetched the page with curl and read the text; saw the byline date and the sentence.
The words: "In the next five years, computer programs that can think will read legal documents and give medical advice."
Coded statement row: altman-2021-01
What it commits to (capability): Within five years of March 2021 (by about March 2026), AI programs that 'can think' would read legal documents and give medical advice.
Due: 2026 ('In the next five years' said 2021-03-16 gives 2026-03-16.)
Today (2026-10-17): hit. Both activities exist at scale. OpenAI's technical report says GPT-4 passed a simulated bar exam with a score around the top 10% of test takers (arXiv 2303.08774, March 2023). OpenAI's own ChatGPT Health page (January 2026) says over 230 million people ask health and wellness questions on ChatGPT every week. Judgment call: 'can think' has no test, and OpenAI's page says Health 'is not intended for diagnosis or treatment', so 'medical advice' is met in the loose sense of answering health questions, not as a licensed service. Evidence: https://openai.com/index/introducing-chatgpt-health/
Flag: loose bar: the claim names no standard
Notes: Other evidence: https://arxiv.org/abs/2303.08774 (read the abstract). The CSV row altman-2021-01 codes this passage as a decades-away statement and does not capture the explicit five-year and ten-year points; the quote itself matches the page.

altman-2021-03-16-ten-years-assembly-line-companions

Said: 2021-03-16 (Byline date on the essay page: 'by Sam Altman · March 16, 2021'.)
Occasion: Essay 'Moore's Law for Everything' on his own site.
Source (own words): https://moores.samaltman.com/, at Same paragraph as the five-year sentence, second sentence.
Checked how: Fetched the page with curl and read the text.
The words: "In the next decade, they will do assembly-line work and maybe even become companions."
Coded statement row: altman-2021-01
What it commits to (deployment): Within ten years of March 2021 (by about March 2031), AI programs would do assembly-line work and maybe even become companions.
Due: 2031 ('In the next decade' said 2021-03-16 gives 2031-03-16.)
Today (2026-10-17): not yet due. The ten years run to March 2031. Not scored.
Notes: The word 'maybe' hedges only the companions half. In the same paragraph he says that 'in the decades after that, they will do almost everything'; that sentence has no date and was not scored.

altman-2023-05-22-ten-years-expert-skill

Said: 2023-05-22 (Date printed on the OpenAI post: 'May 22, 2023'.)
Occasion: OpenAI blog post 'Governance of superintelligence', authors listed as Sam Altman, Greg Brockman and Ilya Sutskever.
Source (own words): https://web.archive.org/web/2023/https://openai.com/blog/governance-of-superintelligence, at First paragraph.
Checked how: openai.com returned 403 to curl. Read the Wayback Machine copy of the OpenAI page and saw the date, the three authors and the sentence.
The words: "it’s conceivable that within the next ten years, AI systems will exceed expert skill level in most domains"
Coded statement row: altman-2023-06
What it commits to (capability): Within ten years of May 2023 (by May 2033), AI systems might exceed expert skill level in most domains and do as much productive work as one of today's largest corporations.
Due: 2033 ('within the next ten years' said 2023-05-22 gives 2033.)
Today (2026-10-17): not yet due. The ten years run to May 2033. Not scored.
Notes: Hedge in his words: 'Given the picture as we see it now, it's conceivable'. The post has three named authors, so the wording is joint, not his alone; the CSV row altman-2023-06 lists it under Altman only. The rest of the sentence reads 'and carry out as much productive activity as one of today's largest corporations'.

altman-2024-03-18-end-of-decade-capable-systems

Said: 2024-03-18 (Lex Fridman's transcript page was posted 2024-03-17 21:13 US Eastern, which is 2024-03-18 01:13 UTC (page metadata datePublished); the page body says March 17. The CSV date of 2024-03-18 is the UTC date.)
Occasion: Lex Fridman Podcast #419, interview with Altman; he says this in the opening clip and again at about 1:32.
Source (transcript): https://lexfridman.com/sam-altman-2-transcript/, at Opening remarks (00:00:00); repeated in the AGI section (about 01:33).
Checked how: Fetched the transcript page with curl and found the sentence at the start and in the AGI section.
The words: "I expect that by the end of this decade, and possibly somewhat sooner than that, we will have quite capable systems"
Coded statement row: altman-2024-02
What it commits to (capability): By the end of the 2020s, possibly sooner, there would be AI systems capable enough that we would look at them and say 'Wow, that's really remarkable.'
Due: 2030 ('by the end of this decade' said in 2024. 2029 or 2030 depending on how the decade is counted; 2030 used as the outer year.)
Today (2026-10-17): not yet due. The decade has not ended. Not scored.
Notes: He frames this as a deliberate non-answer on AGI timing: 'in the interest of not trying to dodge a question'. The 'wow, that's remarkable' test is subjective. A later restatement with a firmer year (Fortune, September 2025, 'by 2030') was not opened as a primary source and is not entered.

altman-2024-03-18-new-model-this-year

Said: 2024-03-18 (Same transcript page and date as the end-of-decade claim; the sentence is at 01:06:30 in the episode.)
Occasion: Lex Fridman Podcast #419, answering 'when is GPT-5 coming out again?'
Source (transcript): https://lexfridman.com/sam-altman-2-transcript/, at GPT-5 section, 01:06:30.
Checked how: Fetched the transcript page with curl and read the exchange.
The words: "We will release an amazing new model this year."
What it commits to (deployment): OpenAI would release an 'amazing' new model during 2024, name not decided.
Due: 2024 ('this year', said in March 2024.)
Today (2026-10-17): hit. OpenAI released o1-preview on 2024-09-12 and described it as a new series of reasoning models. 'Amazing' is a judgment word; the release itself is checkable. Evidence: https://openai.com/index/introducing-openai-o1-preview/
Notes: He added 'I don't know what we'll call it' and, when asked directly about GPT-5, said 'I don't know. That's the honest answer.' Read the OpenAI page through a reader proxy (r.jina.ai) after checking the date shown on it, September 12, 2024. Not in the CSV.

altman-2024-09-23-superintelligence-few-thousand-days

Said: 2024-09-23 (Date printed under the title: 'September 23, 2024'.)
Occasion: Essay 'The Intelligence Age' on his own site.
Source (own words): https://ia.samaltman.com/, at Second paragraph.
Checked how: Fetched the page with curl and read the sentence and date.
The words: "It is possible that we will have superintelligence in a few thousand days (!); it may take longer, but I’m confident we’ll get there."
Coded statement row: altman-2024-03
What it commits to (timing): Superintelligence might arrive in a few thousand days from September 2024, though possibly later.
Due: 2034 (A range, not a point. Reading 'a few thousand days' as 2,000 to 3,000 days gives about March 2030 to June 2032. A looser reading of 1,000 to 3,650 days spans 2027 to 2034. 2034 is used as the outer year.)
Today (2026-10-17): not yet due. Even the earliest reading (1,000 days) falls in June 2027. Not scored.
Notes: Hedges in his words: 'It is possible' and 'it may take longer'. The CSV row altman-2024-03 matches the page. 'Superintelligence' is not defined in the essay.

altman-2025-01-05-agents-join-workforce-2025

Said: 2025-01-05 (Blog Atom feed (https://blog.samaltman.com/posts.atom) timestamp 2025-01-06T01:37:29Z, which is 2025-01-05 in Pacific time. The post page shows no date. The CSV date of 2025-01-06 is the UTC date.)
Occasion: Blog post 'Reflections' on his own site.
Source (own words): https://blog.samaltman.com/reflections, at Paragraph after 'We are now confident we know how to build AGI as we have traditionally understood it.'
Checked how: Fetched the page with curl and read the sentence; took the date from the Atom feed because the page does not print one.
The words: "in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies."
Coded statement row: altman-2025-03
What it commits to (deployment): During 2025 the first AI agents might 'join the workforce' and materially change what companies produce.
Due: 2025 (His words: 'in 2025'.)
Today (2026-10-17): unfalsifiable. Two parts. Agents doing paid work did ship in 2025, and he wrote in June 2025 that '2025 has seen the arrival of agents that can do real cognitive work' (his own assessment in The Gentle Singularity). 'Materially change the output of companies' has no measure, and the evidence is mixed: the Remote Labor Index (arXiv 2510.26787, October 2025) reports that the best agents completed a very low share of real freelance projects at acceptable quality. Combined with the 'may' hedge, there is no clear test. This is a judgment call for John to review. Evidence: https://www.remotelabor.ai/
Notes: I read the Remote Labor Index site, which gives no exact number; McKinsey's 2025 State of AI could not be opened (timed out), so no survey figure is cited. The CSV quote matches the page.

altman-2025-01-05-agi-during-this-presidents-term

Said: 2025-01-05 (Bloomberg page: 'January 5, 2025, Updated on January 6, 2025'.)
Occasion: Bloomberg Businessweek printed interview by Josh Tyrangiel, 'Sam Altman on ChatGPT's First Two Years, Elon Musk and AI Under Trump', edited for clarity and length.
Source (printed interview): https://brp-prod-bcc.bloomberg.com/features/2025-sam-altman-interview/, at Answer about supporting the inauguration and working with the president.
Checked how: Fetched the page with curl and read the answer.
The words: "I think AGI will probably get developed during this president’s term, and getting that right seems really important."
Coded statement row: altman-2025-02
What it commits to (timing): AGI would probably be developed between January 2025 and the end of the current presidential term in January 2029.
Due: 2029 ('this president's term': Trump's term that began 2025-01-20 ends 2029-01-20.)
Today (2026-10-17): not yet due. The term runs to January 2029. Not scored.
Notes: Hedge in his words: 'probably'. In the same interview he said he once thought 'we would do it in 2025, but it was a totally random number based off of 10 years from when we started'. The interview URL is a Bloomberg mirror host (brp-prod-bcc.bloomberg.com) taken from the CSV.

altman-2025-06-10-2026-systems-find-novel-insights

Said: 2025-06-10 (Blog Atom feed timestamp 2025-06-10T21:12:47Z; the post page shows no date.)
Occasion: Blog post 'The Gentle Singularity' on his own site.
Source (own words): https://blog.samaltman.com/the-gentle-singularity, at Paragraph beginning '2025 has seen the arrival of agents...'.
Checked how: Fetched the page with curl and read the sentences; took the date from the Atom feed.
The words: "2026 will likely see the arrival of systems that can figure out novel insights."
What it commits to (capability): In 2026 there would probably be AI systems that can figure out novel insights.
Due: 2026 (His words: '2026'.)
Today (2026-10-17): hit. On 2026-05-20 OpenAI announced that an internal model had disproved a long-standing conjecture on the planar unit distance problem (first posed by Erdős in 1946), and said the proof was checked by external mathematicians. Scientific American (2026-05-21) names outside people who checked or commented, including Daniel Litt and Tim Gowers, and reports caveats: humans did substantial clean-up, and the mathematical tools used were not new. The main evidence is OpenAI's own announcement plus outside checking. Evidence: https://openai.com/index/model-disproves-discrete-geometry-conjecture/
Notes: Hedge in his words: 'likely'. 'Novel insights' is not defined; I judged a verified new mathematical result to meet it. Scientific American: https://www.scientificamerican.com/article/ai-just-solved-an-80-year-old-erdos-problem-and-mathematicians-are-amazed/. The OpenAI page returned 403 to WebFetch and was read through a reader proxy; its date comes from search results and the Scientific American piece ('the day before'). Existing CSV rows altman-2025-06a and -06b use a different sentence from this post.

altman-2025-06-10-2027-robots-real-world-tasks

Said: 2025-06-10 (Blog Atom feed timestamp 2025-06-10T21:12:47Z; the post page shows no date.)
Occasion: Blog post 'The Gentle Singularity' on his own site.
Source (own words): https://blog.samaltman.com/the-gentle-singularity, at Same paragraph as the 2026 sentence, next sentence.
Checked how: Fetched the page with curl and read the sentence; took the date from the Atom feed.
The words: "2027 may see the arrival of robots that can do tasks in the real world."
What it commits to (capability): In 2027 there might be robots that can do tasks in the real world.
Due: 2027 (His words: '2027'.)
Today (2026-10-17): not yet due. 2027 has not started. Not scored.
Notes: Hedge in his words: 'may'. 'Tasks in the real world' has no stated difficulty or scale.

altman-2025-10-14-erotica-for-verified-adults-in-december

Said: 2025-10-14 (Post timestamp 'Tue Oct 14 16:02:42 +0000 2025' (UTC).)
Occasion: X post about relaxing ChatGPT's restrictions.
Source (own words): https://x.com/sama/status/1978129344598827128, at Last paragraph of the post.
Checked how: x.com not fetched directly. Read the post text and timestamp through the fxtwitter mirror (https://api.fxtwitter.com/sama/status/1978129344598827128).
The words: "In December, as we roll out age-gating more fully ... we will allow even more, like erotica for verified adults."
What it commits to (deployment): In December 2025 ChatGPT would allow more content, including erotica, for age-verified adults.
Due: 2025 (His words: 'In December' (2025).)
Today (2026-10-17): missed. Coverage reports that the launch slipped from December to the first quarter of 2026 (Fidji Simo, OpenAI's head of applications, on 2025-12-11), was delayed again on 2026-03-06, and was paused indefinitely around 2026-03-26. A page last updated 2026-09-26 says it had not launched and had no date. I did not open an OpenAI statement; all of this is news coverage, citing the Financial Times and an OpenAI spokesperson. Evidence: https://www.thurrott.com/a-i/334240/openai-is-no-longer-working-on-adult-mode-for-chatgpt
Flag: the verdict rests on press coverage; a primary source for the slip is still to be found
Notes: Quote has an ellipsis; the omitted words are 'and as part of our “treat adult users like adults” principle,'. Other pages: https://justainews.com/blog/adult-mode-in-chatgpt-explained-nsfw-erotica-porn-policy/ (September 2026 status). Axios (https://www.axios.com/2026/03/06/openai-delays-chatgpt-adult-mode) returned 403 and was not read. Not in the CSV.

altman-2025-10-29-automated-research-intern-by-september-2026

Said: 2025-10-29 (Post timestamp 'Wed Oct 29 17:19:00 +0000 2025' (UTC); the post says the livestream was 'Yesterday'.)
Occasion: X post summarising OpenAI's livestream of 2025-10-28.
Source (own words): https://x.com/sama/status/1983584366547829073, at First paragraph after 'TL;DR:'.
Checked how: x.com not fetched directly. Read the post through the fxtwitter mirror (https://api.fxtwitter.com/sama/status/1983584366547829073).
The words: "We have set internal goals of having an automated AI research intern by September of 2026 running on hundreds of thousands of GPUs"
Coded statement row: altman-2025-09
What it commits to (capability): OpenAI's internal goal: by September 2026, an automated AI research intern running on hundreds of thousands of GPUs.
Due: 2026 (His words: 'by September of 2026'.)
Today (2026-10-17): hit. ONLY THE COMPANY'S OWN CLAIM. OpenAI's post of 2026-09-06 says: 'According to our measurements, we have now reached the goal, announced last fall, of having an automated research intern by September of this year.' It defines the intern as 'a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days.' I found no independent check. The post, as I read it, does not say whether it ran on hundreds of thousands of GPUs, which was part of the original wording. TIME (2026-08-26) reports chief scientist Jakub Pachocki saying the company had already met its internal benchmark; that is also the company's own word. Evidence: https://openai.com/index/research-acceleration-view-inside-openai/
Flag: the evidence is OpenAI's own word only
Notes: Hedge in his words: 'We may totally fail at this goal'. It is a goal, not a forecast. OpenAI's page returned 403 to WebFetch and curl and was read in full through a reader proxy (r.jina.ai). Its date of 2026-09-06 comes from search results and Simon Willison's post (https://simonwillison.net/2026/Sep/6/research-acceleration-the-view-inside-openai/). The same page says people still set research priorities and that it paused some training after the Hugging Face incident. The CSV quote for altman-2025-09 matches the post.

altman-2025-10-29-automated-ai-researcher-by-march-2028

Said: 2025-10-29 (Post timestamp 'Wed Oct 29 17:19:00 +0000 2025' (UTC).)
Occasion: X post summarising OpenAI's livestream of 2025-10-28.
Source (own words): https://x.com/sama/status/1983584366547829073, at First paragraph after 'TL;DR:', second half of the sentence.
Checked how: Read through the fxtwitter mirror (https://api.fxtwitter.com/sama/status/1983584366547829073).
The words: "a true automated AI researcher by March of 2028. We may totally fail at this goal"
Coded statement row: altman-2025-09
What it commits to (capability): OpenAI's internal goal: by March 2028, a true automated AI researcher.
Due: 2028 (His words: 'by March of 2028'.)
Today (2026-10-17): not yet due. March 2028 has not arrived. OpenAI's 2026-09-06 post says 'We are making strong progress toward creating an automated AI researcher by March of 2028'; that is the company's own statement, not a result. Evidence: https://openai.com/index/research-acceleration-view-inside-openai/
Notes: The post also says OpenAI does not know how to get to full recursive self-improvement safely. Hedge: 'We may totally fail at this goal'.

altman-2025-10-29-small-new-discoveries-in-2026

Said: 2025-10-29 (Post timestamp 'Wed Oct 29 17:19:00 +0000 2025' (UTC).)
Occasion: X post summarising OpenAI's livestream of 2025-10-28.
Source (own words): https://x.com/sama/status/1983584366547829073, at Last paragraph of the post.
Checked how: Read through the fxtwitter mirror (https://api.fxtwitter.com/sama/status/1983584366547829073).
The words: "In 2026 we expect that our AI systems may be able to make small new discoveries; in 2028 we could be looking at big ones."
What it commits to (capability): In 2026 OpenAI's AI systems might be able to make small new discoveries; in 2028 they could make big ones.
Due: 2026 (His words: 'In 2026'. The second half ('in 2028 ... big ones') is a separate 2028 prediction, not scored here.)
Today (2026-10-17): hit. Same evidence as the June 2025 claim: OpenAI's 2026-05-20 announcement that an internal model disproved the Erdős unit distance conjecture, checked by outside mathematicians. Arguably larger than 'small', which still meets a 'may be able to make small' claim. The 2028 half is not yet due. Evidence: https://openai.com/index/model-disproves-discrete-geometry-conjecture/
Restates: altman-2025-06-10-2026-systems-find-novel-insights
Notes: RESTATEMENT of the 2026 sentence in 'The Gentle Singularity', in narrower and more hedged words ('small new discoveries', 'may be able to'). Hedges in his words: 'we expect', 'may', 'could'. Same source post as altman-2025-09 in the CSV, but a different sentence, so no CSV id is given.

altman-2026-08-26-internal-system-called-agi-by-end-of-year

Said: 2026-08-26 (TIME page metadata: published 2026-08-26T11:00:04Z.)
Occasion: TIME profile 'Inside OpenAI's Reboot' by Alex Heath; a reporter's article, not a printed question-and-answer interview.
Source (coverage only): https://time.com/article/2026/08/26/openai-sam-altman-interview/, at Paragraph beginning 'Amid all this, company leaders believe...'.
Checked how: time.com returned an empty body to curl; read the article through a reader proxy (r.jina.ai). The sentence is the reporter's paraphrase of what Altman told him; only the two words 'not quite yet' are given as his. No transcript or Altman post found.
The words: none: coverage only
Coded statement row: altman-2026-04
What it commits to (timing): Per TIME's reporter: by the end of 2026 OpenAI would have an internal system that Altman would call AGI. At the time he said OpenAI was 'not quite yet' there.
Due: 2026 (Reporter's words: 'by the end of the year' (2026).)
Today (2026-10-17): not yet due. The year has not ended. Not scored. Note that the test is that Altman himself would call it AGI, so the answer will rest on his word.
Notes: The CSV row altman-2026-04 quotes the reporter's sentence as if it were Altman's; it is not his own words. OpenAI's charter definition of AGI is quoted by TIME as 'highly autonomous systems that outperform humans at most economically valuable work.' He is directly quoted elsewhere in the article saying of a new model 'I expect this will be the first model where the model actually invents new things in a way that matters', undated and not scored.

Dario Amodei

amodei-2023-07-25-senate-bio-two-to-three-years

Said: 2023-07-25 (Hearing date from the Senate Judiciary Committee hearing page ("Tuesday, July 25th, 2023"). The PDF file name says 07-26, but the hearing page gives the 25th.)
Occasion: Written opening statement to the Senate Judiciary subcommittee on privacy, technology and the law, hearing "Oversight of A.I.: Principles for Regulation"
Source (own words): https://www.judiciary.senate.gov/imo/media/doc/2023-07-26_-_testimony_-_amodei.pdf, at Section on biological risk, sentence beginning "However, a straightforward extrapolation..."; the same statement also has a paragraph "Short-Term, Medium-Term, and Long-Term Risks" that says medium-term risks are those faced in "two to three years"
Checked how: Downloaded the PDF with curl, pulled its text out with the macOS PDF reader (PDFKit) and read the sentences. I did not watch or read the spoken version.
The words: "a straightforward extrapolation of today’s systems to those we expect to see in 2-3 years suggests a substantial risk"
What it commits to (capability): In 2023 he said that in two to three years (roughly mid-2025 to mid-2026) AI systems could become able to fill in all the missing steps needed for a large-scale biological attack, if appropriate guardrails and safeguards were not put in place. The same statement says that by then AI systems may become much better at science and engineering, and could be misused on a grand scale in cyber-security, nuclear technology, chemistry and especially biology.
Due: 2026 ("2-3 years" counted from July 2023, so the window ends around July 2026.)
Today (2026-10-17): unfalsifiable. The claim is conditional ("if appropriate guardrails and mitigations are not put in place") and speaks of a "substantial risk", not of a certainty. Guardrails were put in place, so the case it describes never happened. The nearest outcome evidence is Anthropic's own post of 2025-05-22: it switched on its stricter bio-weapons protections for Claude Opus 4 because it said it could no longer rule out the risk "in the way it was for every previous model". That is the company's own statement and not an outside test. Evidence: https://www.anthropic.com/news/activating-asl3-protections
Notes: Repeated on the Dwarkesh podcast on 2023-08-08 ("two or three years from now, we're going to have a real problem"). In "The Adolescence of Technology" (2026-01-26) he writes he is concerned that models "are approaching (or may already have reached)" the knowledge needed to create and release biological weapons end to end.

amodei-2023-08-08-dwarkesh-well-educated-human-two-three-years

Said: 2023-08-08 (Date shown on the episode page ("Aug 08, 2023").)
Occasion: Dwarkesh Podcast interview (episode now titled "The hidden pattern behind every AI breakthrough")
Source (transcript): https://www.dwarkesh.com/p/dario-amodei, at Transcript timestamp 00:27:56 to 00:29:32, answer to "what does your estimate of when we get something kind of human level look like?"
Checked how: Fetched the episode page with curl and searched its transcript text for the passage.
The words: "generally well educated human, that could be not very far away at all. I think that could happen in two or three years."
What it commits to (timing): In 2023 he said a model that, after an hour of talking to it, seems basically like a generally well educated human could be not very far away, perhaps two or three years (roughly 2025-2026). He added that the main thing that would stop it is the industry or governments slowing down for safety reasons.
Due: 2026 ("two or three years" counted from August 2023, so the window ends around August 2026.)
Today (2026-10-17): unfalsifiable. He gives no test. "Basically like a generally well educated human" after an hour of conversation is a judgement, and I found no study that used that test. It is also conditional on safety limits not slowing the industry.
Notes: Same interview, same time horizon for a bio-weapons risk (see the Senate testimony row). His later "country of geniuses in a datacenter" is a much higher bar than this one.

amodei-2024-10-11-mlg-powerful-ai-as-early-as-2026

Said: 2024-10-11 (The essay page says only "October 2024". The day comes from decoding the ID of Dario Amodei's own X post announcing the essay (2024-10-11 20:00 UTC); I saw that post only in a search result and did not open it.)
Occasion: Essay "Machines of Loving Grace" on his own site
Source (own words): https://darioamodei.com/essay/machines-of-loving-grace, at Introduction, paragraph on when powerful AI could arrive; definition of powerful AI and the "country of geniuses in a datacenter" phrase are in the next section
Checked how: Fetched the page with curl, stripped the markup and read the passage.
The words: "I think it could come as early as 2026, though there are also ways it could take much longer."
What it commits to (timing): "Powerful AI" (his term, defined in the essay as an AI smarter than a Nobel Prize winner across most relevant fields, which he sums up as "a country of geniuses in a datacenter") could be built as early as 2026, though it could take much longer.
Due: 2026 ("as early as 2026" is an earliest-case date, stated with "could" and "though there are also ways it could take much longer".)
Today (2026-10-17): not yet due. 2026 is not over and the claim is hedged. His own essay of 2026-06-10 says that if the scaling trend "continues for only a year or two longer" we are "likely" to get Powerful AI, so he did not say it had arrived by June. Evidence: https://www.darioamodei.com/post/policy-on-the-ai-exponential
Notes: This is the earliest of his "2026" statements. Restated on the Lex Fridman podcast 2024-11-11 ("we'll get there by 2026 or 2027"), in the DeepSeek essay 2025-01-29, in "The Urgency of Interpretability" 2025-04-24, and at Davos 2026-01-20 when the moderator asked if he stood by "2026 or 2027" and he said "I don't think that's going to turn out to be that far off". The essay also says powerful AI could compress 50-100 years of biology progress into 5-10 years after it arrives; that claim has no calendar date, so it is not listed separately.

amodei-2024-11-11-lex-asl3-next-year

Said: 2024-11-11 (Date shown on the transcript page ("November 11, 2024").)
Occasion: Lex Fridman Podcast #452
Source (transcript): https://lexfridman.com/dario-amodei-transcript, at Section "ASL-3 and ASL-4", timestamp 01:05:44
Checked how: Fetched the transcript page with curl and searched the text.
The words: "I would not be surprised at all if we hit ASL-3 next year."
What it commits to (capability): Anthropic could reach "ASL-3" in 2025 (it might even have happened in 2024). ASL-3 is the level of risk at which Anthropic's own safety policy requires stricter security and deployment protections, here mainly about help with chemical, biological, radiological or nuclear weapons. He added he would be "very, very surprised" if it were 2030.
Due: 2025 ("next year", said in November 2024.)
Today (2026-10-17): hit. Anthropic's own post of 2025-05-22 says it activated the ASL-3 protections with the launch of Claude Opus 4. Caveat in the same post: it said it had "not yet determined" that the model definitely passed the threshold, and switched the protections on as a precaution because it could no longer clearly rule the risk out. The only evidence is Anthropic's own statement. Evidence: https://www.anthropic.com/news/activating-asl3-protections
Flag: the evidence is Anthropic's own post, which says protections were switched on as a precaution
Notes: Hedged in his words ("would not be surprised"). Counted as a hit on the plain reading that Anthropic reached ASL-3 protections during 2025.

amodei-2025-01-29-deepseek-essay-smarter-than-almost-all-humans-2026-2027

Said: 2025-01-29 (The page says "January 2025". The day comes from the first Hacker News submission of the page (2025-01-29 16:19 UTC).)
Occasion: Essay "On DeepSeek and Export Controls" on his own site
Source (own words): https://darioamodei.com/on-deepseek-and-export-controls, at Section on what a world with powerful AI by 2026-2027 looks like; the sentence begins "Making AI that is smarter than almost all humans at almost all things..."
Checked how: Fetched the page with curl and searched the text. The quote is the end of a longer sentence whose first half reads "Making AI that is smarter than almost all humans at almost all things will require millions of chips, tens of billions of dollars (at least),".
The words: "and is most likely to happen in 2026-2027"
What it commits to (timing): AI that is smarter than almost all humans at almost all things is most likely to be built in 2026-2027.
Due: 2027 (His own range "2026-2027".)
Today (2026-10-17): not yet due. 2027 has not been reached. His 2026-06-10 essay still talks of "a year or two longer" for Powerful AI. Evidence: https://www.darioamodei.com/post/policy-on-the-ai-exponential
Restates: amodei-2024-10-11-mlg-powerful-ai-as-early-as-2026
Notes: Firmer than the October 2024 wording ("most likely" instead of "could") and a different bar ("smarter than almost all humans at almost all things" rather than "smarter than a Nobel Prize winner").

amodei-2025-03-10-cfr-ninety-percent-of-code-in-three-to-six-months

Said: 2025-03-10 (Event date on the Council on Foreign Relations page: "Monday March 10, 2025 6:00 p.m. (EDT)". Some news outlets say March 11.)
Occasion: Council on Foreign Relations "CEO Speaker Series", conversation with CFR president Michael Froman
Source (transcript): https://www.cfr.org/event/ceo-speaker-series-dario-amodei-anthropic, at Answer about coding, roughly the middle of the transcript; the page marks it "an uncorrected transcript"
Checked how: Fetched the CFR page with curl; the transcript is printed on the page. I read the passage in the text.
The words: "I think we’ll be there in three to six months—where AI is writing 90 percent of the code."
What it commits to (deployment): By about June to September 2025, AI would be writing 90 percent of the code. He did not say whose code; the sentence is "we are not far from the world" and the next sentence adds that within twelve months AI may write essentially all of it (see the next row).
Due: 2025 ("three to six months" counted from 2025-03-10, so by about 2025-09-10.)
Today (2026-10-17): missed. Judgement call, flagged for John. Read as written (code in general), nothing shows 90 percent: I found no reliable measure for all code, and the best outside estimate (Redwood Research, 2025-10-22) put AI-written lines of code merged at Anthropic at around 50 percent on average, with some teams near 90. Amodei said at Dreamforce in October 2025 that it was "absolutely true" inside Anthropic and some partners, then "on many teams, not uniformly". On 2026-02-13 he said it "happened, at least at some places" and called the 90 percent test "a very weak criterion". Read as "at Anthropic", the only evidence is the company's own statements. Evidence: https://www.dwarkesh.com/p/dario-amodei-2
Flag: contested: whose code is unstated, and only company statements bear on it
Notes: Dreamforce wording as transcribed by Redwood (https://blog.redwoodresearch.org/p/is-90-of-code-at-anthropic-being), not opened as a recording. In the 2026-02-13 interview he also says people wrongly took the claim to mean 90 percent fewer software engineers.

amodei-2025-03-10-cfr-essentially-all-code-in-twelve-months

Said: 2025-03-10 (Same event and date as the previous row (CFR page: "Monday March 10, 2025").)
Occasion: Council on Foreign Relations "CEO Speaker Series", conversation with Michael Froman
Source (transcript): https://www.cfr.org/event/ceo-speaker-series-dario-amodei-anthropic, at Same answer as the previous row, the sentence after the 90 percent sentence
Checked how: Fetched the CFR page with curl and read the passage in the transcript text.
The words: "in twelve months, we may be in a world where AI is writing essentially all of the code."
What it commits to (deployment): By about March 2026, AI would be writing essentially all of the code. He added that the programmer would still need to specify what is being built.
Due: 2026 ("twelve months" counted from 2025-03-10, so by about 2026-03-10.)
Today (2026-10-17): missed. Judgement call, flagged for John. On 2026-02-13 he listed "100% of code is written by the model" as a further step on a spectrum and put full end-to-end software engineering "one or two years" out; he did not say the 100 percent step had been reached. No outside measure shows essentially all code is written by AI. Company-side statements (coverage only, not opened) say Anthropic's own code was nearly all written by Claude by February 2026. Evidence: https://www.dwarkesh.com/p/dario-amodei-2
Flag: contested: whose code is unstated, and only company statements bear on it
Notes: Carries the same ambiguity as the 90 percent row: he did not say whose code.

amodei-2025-05-28-axios-half-of-entry-level-white-collar-jobs

Said: 2025-05-28 (Date taken from the Axios article's web address (2025/05/28). I could not open the article.)
Occasion: Interview with Axios ("Behind the Curtain: A white-collar bloodbath")
Source (coverage only): https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic, at Axios article of 2025-05-28; interview at his San Francisco office
Checked how: The Axios page returned HTTP 403 to curl, so I did not read it. The wording of the claim was confirmed in a later primary transcript: on 60 Minutes (2025-11-16, https://www.cbsnews.com/news/anthropic-ceo-dario-amodei-warning-of-ai-potential-dangers-60-minutes-transcript/) the interviewer read it back as "AI could wipe out half of all entry-level white-collar jobs and spike unemployment to 10% to 20% in the next one to five years" and Amodei answered "Yes". No quote is given because the original source was not opened.
The words: none: coverage only
What it commits to (deployment): AI could eliminate half of all entry-level white-collar jobs and push US unemployment to 10-20 percent within one to five years.
Due: 2030 ("next one to five years" counted from May 2025, so the window ends in May 2030.)
Today (2026-10-17): not yet due. The window runs to 2030. For context only: the US unemployment rate was 4.2 percent in September 2026 (Bureau of Labor Statistics release of 2026-10-02), and an NPR report of 2026-08-18 gives the New York Fed rate for recent college graduates as 5.7 percent in June 2026 against 4.1 percent for all workers, with the Fed's own analysis not putting the rise down to AI (NPR summary, not opened). Evidence: https://www.bls.gov/news.release/archives/empsit_10022026.htm
Notes: Restated in his own words in "The Adolescence of Technology" (2026-01-26): "I predicted that AI could displace half of all entry-level white collar jobs in the next 1-5 years". Search coverage dated 2026-05-19 says he repeated the warning on Fox News (not opened). A third-party tracker claims he reframed this in May 2026 as job-expanding; I could not find support for that.

amodei-2026-01-20-davos-software-engineers-end-to-end-six-to-twelve-months

Said: 2026-01-20 (Session "The Day After AGI" at the World Economic Forum in Davos. The day is inferred: a WEF "Radio Davos day 2" episode and a third-party tracker point to 2026-01-20; a Diginomica article is dated 2026-01-21. I could not open the WEF page (HTTP 403).)
Occasion: World Economic Forum Annual Meeting 2026, on-stage conversation with Demis Hassabis moderated by Zanny Minton Beddoes of The Economist
Source (transcript): https://aletteraday.substack.com/p/letters-314315-demis-hassabis-and, at "Transcript - 2026 Talk", his answer to the first question (whether he still stands by his "2026 or 2027" timeline)
Checked how: Fetched the newsletter page with curl and read the transcript text. The host is a third-party newsletter (posted 2026-01-27); the official WEF transcript at https://www.weforum.org/podcasts/radio-davos/episodes/ai-agi-dario-amodei-demis-hassabis/ returned HTTP 403 so I could not compare the two.
The words: "We might be six to twelve months away from when the model is doing most, maybe all, of what software engineers do end-to-end."
What it commits to (capability): About six to twelve months after January 2026 (July 2026 to January 2027), AI models would be doing most, maybe all, of what software engineers do, from start to finish.
Due: 2027 ("six to twelve months" counted from 2026-01-20, so the window ends around 2027-01-20.)
Today (2026-10-17): not yet due. The window runs to January 2027. On 2026-02-13 he gave a longer span for end-to-end coding: "I think we'll be there in one or two years" (apart from irreducible uncertainty). Evidence: https://www.dwarkesh.com/p/dario-amodei-2
Notes: A newer, end-to-end version of the March 2025 code claims. In the same answer he said engineers inside Anthropic told him "I don't write any code anymore", which is a company statement.

amodei-2026-01-26-adolescence-powerful-ai-one-to-two-years

Said: 2026-01-26 (The essay page says "January 2026". The day comes from the ID of his own X post about it (2026-01-26 17:03 UTC, seen only in a search result) and the first Hacker News submission (2026-01-26 17:07 UTC).)
Occasion: Essay "The Adolescence of Technology" on his own site
Source (own words): https://www.darioamodei.com/essay/the-adolescence-of-technology, at Section on the powerful-AI timeline; the sentence begins "As I wrote in Machines of Loving Grace..."
Checked how: Fetched the page with curl, stripped the markup and read the passage.
The words: "powerful AI could be as little as 1–2 years away, although it could also be considerably further out."
What it commits to (timing): Powerful AI (a "country of geniuses in a datacenter") could be as little as one to two years away, that is by about January 2028, though it could be considerably further out.
Due: 2028 ("1-2 years" counted from January 2026.)
Today (2026-10-17): not yet due. 2028 has not been reached. His 2026-06-10 essay says: "If these scaling laws continue for only a year or two longer, we are likely to get what I've called Powerful AI". Evidence: https://www.darioamodei.com/post/policy-on-the-ai-exponential
Restates: amodei-2024-10-11-mlg-powerful-ai-as-early-as-2026
Notes: He also writes there that the loop in which today's AI builds the next generation "may be only 1-2 years away" from fully closing, and poses a scenario of a "country of geniuses" appearing "in ~2027". The scenario is a thought experiment, not a forecast.

amodei-2026-02-13-dwarkesh-trillions-of-dollars-revenue-before-2030

Said: 2026-02-13 (Date shown on the episode page ("Feb 13, 2026").)
Occasion: Dwarkesh Podcast, episode "We are near the end of the exponential"
Source (transcript): https://www.dwarkesh.com/p/dario-amodei-2, at Discussion of compute spending and the pace of revenue; he introduces it with "So I'll make another prediction."
Checked how: Fetched the episode page with curl and searched its transcript text.
The words: "It is hard for me to see that there won’t be trillions of dollars in revenue before 2030."
What it commits to (deployment): Revenue in the AI industry (the transcript does not define it more closely) will reach trillions of dollars before 2030. He gave a slower case of "maybe three years" in which the real "country of geniuses" arrives in 2028 and revenue goes from the low hundreds of billions in 2028 to trillions two years later.
Due: 2029 ("before 2030", read as by the end of 2029.)
Today (2026-10-17): not yet due. The deadline is the end of 2029. The slower case he describes would put trillions at about 2030, so the claim is not clearly settled even then.
Notes: Same interview: country of geniuses "in one to two years, maybe one to three" (a "50/50 hunch"), 90 percent confidence within ten years, and "crazy to say that this won't happen by 2035". He also says Anthropic's revenue went from about $1 billion to $9-10 billion in 2025; that is the company's own figure.

amodei-2026-09-12-pace-the-frontier-swarm-six-to-twelve-months

Said: 2026-09-12 (The page says "September 2026". The day comes from the first Hacker News submission (2026-09-12 14:10 UTC); Axios and TechCrunch web addresses (seen in search results, not opened) also carry 2026/09/12.)
Occasion: Essay "We Must Pace the Frontier" on his own site
Source (own words): https://www.darioamodei.com/post/we-must-pace-the-frontier, at Paragraph about his second concern, the "OpenAI-Hugging Face incident"
Checked how: Fetched the page with curl, stripped the markup and read the passage.
The words: "it’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet"
What it commits to (capability): He worries that, in six to twelve months (by about September 2027), a swarm of AI agents with a similar fault to the one in the incident he describes could take over the entire internet with a persistent botnet, causing potentially hundreds of billions of dollars of damage.
Due: 2027 ("6-12 months" counted from September 2026.)
Today (2026-10-17): not yet due. The window runs to about September 2027. It is stated as a worry ("it's my worry that ... could"), not a flat forecast.
Notes: The "OpenAI-Hugging Face incident" is described in the essay; I did not check it independently. The essay also says AI "could cure most major diseases in the next 5-10 years", a restatement of the 2024 essay with a start date.

Jensen Huang

huang-2023-11-29-dealbook-ai-fairly-competitive-with-humans-five-years

Said: 2023-11-29 (New York Times DealBook Summit 2023. News coverage (NBC) says the remarks were on a Wednesday, which is 2023-11-29; the video was uploaded to YouTube on 2023-11-30.)
Occasion: On-stage interview with Andrew Ross Sorkin at the New York Times DealBook Summit
Source (transcript): https://www.youtube.com/watch?v=Pkj-BLHs6dE, at Video "Jensen Huang of Nvidia on the Future of A.I.", about 00:07:21 to 00:08:06; answer to "Do you think in 10 years from now we are there?"
Checked how: Downloaded the English closed captions of the New York Times Events video with yt-dlp and read them. The captions are machine-made and contain errors ("a GI", "a eyes"), so the quote reproduces them as printed.
The words: "deliver results that are fairly fairly competitive to a Normal human, I would I would say that within the next five years."
What it commits to (timing): Asked whether AGI would exist in ten years, he said yes depending on the definition. If AGI means software that can pass a set of basic intelligence tests with results fairly competitive with a normal human, he said you will see AIs that can do that within five years.
Due: 2028 ("within the next five years" from November 2023.)
Today (2026-10-17): not yet due. The window ends around November 2028. Note that on 2026-03-23 he said AGI had already been achieved, on a different definition (see the Lex Fridman row).
Notes: The definition is his own and is about passing tests, not about doing a person's job. Restated at Stanford on 2024-03-01.

huang-2024-03-01-siepr-ai-passes-human-tests-five-years

Said: 2024-03-01 (Reuters (published 2024-03-01) says Huang spoke "on Friday" at an economic forum at Stanford; the SIEPR video was uploaded to YouTube on 2024-03-07.)
Occasion: Keynote conversation with John Shoven at the 2024 Stanford Institute for Economic Policy Research (SIEPR) Economic Summit
Source (transcript): https://www.youtube.com/watch?v=cEg8cOx7UZk, at Video "Keynote by NVIDIA CEO Jensen Huang at 2024 SIEPR Economic Summit", about 00:27:54 to 00:33:00
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made, so punctuation is missing). Date checked against the Reuters report on theglobeandmail.com.
The words: "I'm guessing in 5 years time we'll do well on every single one of them"
What it commits to (timing): If you gave an AI every test you can imagine (math, reasoning, history, biology, medical exams, bar exams), in five years it would do well on every one. So if AGI means passing human tests, he said it is five years away; if it means having human intelligence, he said nobody knows how to specify that, so "we're not sure".
Due: 2029 ("in 5 years time" from March 2024.)
Today (2026-10-17): not yet due. The window ends around March 2029.
Restates: huang-2023-11-29-dealbook-ai-fairly-competitive-with-humans-five-years
Notes: Same test-passing definition as the DealBook answer, with a new date. The question put to him was whether AGI is "50 years away" or "five years away".

huang-2025-03-18-gtc-vera-rubin-second-half-2026

Said: 2025-03-18 (Nvidia GTC 2025 keynote; the NVIDIA channel video was released 2025-03-18.)
Occasion: GTC 2025 keynote, San Jose
Source (transcript): https://www.youtube.com/watch?v=_waPvOwL9Z8, at Roadmap section, about 01:27:00 to 01:28:30 of the video
Checked how: Downloaded the English closed captions with yt-dlp and read them. The captions are machine-made and mis-hear the product name ("Vera ruin mvlink 144" for Vera Rubin NVLink 144); the quote is printed as captioned.
The words: "Vera ruin mvlink 144 is the second half of next year"
What it commits to (deployment): Nvidia's next-generation rack system, Vera Rubin NVLink 144, would come out in the second half of 2026.
Due: 2026 ("the second half of next year", said in March 2025.)
Today (2026-10-17): hit. Nvidia's own results release of 2026-08-26 says the Vera Rubin platform is "ramping into full production" with racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. That is Nvidia's own statement; I did not find an independent audit. I did not check whether the rack is still sold under the name "NVLink 144". Evidence: https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027
Flag: the evidence is Nvidia's own results release
Notes: In the same keynote he said the Blackwell Ultra upgrade comes "second half of this year" (2025); I did not check that one.

huang-2025-03-18-gtc-all-nvidia-software-engineers-ai-assisted-by-end-of-2025

Said: 2025-03-18 (Nvidia GTC 2025 keynote; video released 2025-03-18.)
Occasion: GTC 2025 keynote, San Jose
Source (transcript): https://www.youtube.com/watch?v=_waPvOwL9Z8, at Section on AI agents and the new DGX Spark desktop machine, about 01:45:50 to 01:46:10
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made).
The words: "100% of Nvidia software Engineers will be AI assisted by the end of this year"
What it commits to (deployment): By the end of 2025, every Nvidia software engineer would be assisted by AI.
Due: 2025 ("by the end of this year", said in March 2025.)
Today (2026-10-17): hit. The only evidence is Nvidia's own statement. In the Computex keynote on 2025-05-19 he said "A 100% of NVIDIA software engineers now have digital agents working with them", two months later. No outside check exists. "AI assisted" is a loose test. Evidence: https://www.youtube.com/watch?v=TLzna9__DnI
Flag: the evidence is Nvidia's own word only; "AI assisted" is a loose bar
Notes: Said in the same breath as "100% of them [the world's 30 million software engineers] are going to be AI assisted", which has no date.

huang-2025-03-18-gtc-fifty-million-workers-short-by-end-of-decade

Said: 2025-03-18 (Nvidia GTC 2025 keynote; video released 2025-03-18.)
Occasion: GTC 2025 keynote, San Jose
Source (transcript): https://www.youtube.com/watch?v=_waPvOwL9Z8, at Start of the robotics section, about 01:56:05
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made).
The words: "by the end of this decade the world is going to be at least 50 million workers short"
What it commits to (deployment): By the end of the 2020s the world will be at least 50 million workers short, so robots will have to fill the gap; he said robots might be paid about $50,000 a year.
Due: 2030 ("by the end of this decade", read as 2029 or 2030.)
Today (2026-10-17): not yet due. The deadline is about 2030.
Notes: Restated with a different number at Computex on 2025-05-19: "We have a shortage of workers by 2030 by about 30 to 50 million shortage" (from "at least 50 million" to "30 to 50 million"). A labour-market forecast, included because it is his stated reason for robots.

huang-2025-05-19-computex-mercedes-self-driving-stack-deploying-this-year

Said: 2025-05-19 (Computex 2025 keynote in Taipei; the NVIDIA channel video was released 2025-05-19.)
Occasion: Computex 2025 keynote, Taipei
Source (transcript): https://www.youtube.com/watch?v=TLzna9__DnI, at Robotics and self-driving section, about 01:21:50 to 01:22:15
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made).
The words: "this year we're deploying Mercedes around the world our self-driving car stack end to end stack."
What it commits to (deployment): In 2025 Nvidia would deploy its complete self-driving software stack in Mercedes cars around the world.
Due: 2025 ("this year", said in May 2025.)
Today (2026-10-17): missed. Judgement call, flagged for John: the wording is loose ("deploying", "around the world"). Nvidia's own blog post of 2026-01-05 says its driver-assistance software is coming to US roads "expected by end of this year" (2026), starting with the Mercedes-Benz CLA, and calls it a production launch in the US "this year". Search coverage (not opened) says the same system was already offered in China from late 2025, so a partial rollout may have happened in 2025. Evidence: https://blogs.nvidia.com/blog/drive-av-software-mercedes-benz-cla/
Flag: contested: loose wording ("deploying", "around the world")
Notes: At CES on 2025-01-06 he said a Mercedes fleet is "coming with Nvidia starting with this starting this year going to production" (captions).

huang-2025-10-28-gtc-dc-half-a-trillion-dollars-blackwell-rubin-through-2026

Said: 2025-10-28 (Nvidia GTC Washington, D.C. keynote; the NVIDIA channel video was released 2025-10-28.)
Occasion: GTC Washington, D.C. keynote
Source (transcript): https://www.youtube.com/watch?v=lQHK61IDFH4, at Section on Blackwell and Rubin demand, about 01:02:15
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made).
The words: "visibility into half a trillion dollars of cumulative Blackwell and early ramps of Rubin through 2026."
What it commits to (deployment): Nvidia had orders and visibility for about half a trillion dollars (US$500 billion) of its Blackwell chips and the early ramp of Rubin, covering the five quarters through the end of 2026, excluding China.
Due: 2026 ("through 2026".)
Today (2026-10-17): not yet due. Calendar 2026 is not over. On 2026-03-16 he said he now sees at least $1 trillion through 2027 (next row). Checking either figure needs Nvidia's own filings.
Notes: This is a statement about orders in hand, which only Nvidia can confirm.

huang-2026-03-16-gtc-at-least-one-trillion-dollars-through-2027

Said: 2026-03-16 (Nvidia GTC 2026 keynote; the NVIDIA channel video was released 2026-03-16.)
Occasion: GTC 2026 keynote, San Jose
Source (transcript): https://www.youtube.com/watch?v=jw_o0xr8MWU, at About 00:53:42, right after he recalls the $500 billion figure
Checked how: Downloaded the English closed captions with yt-dlp and read them (machine-made).
The words: "I see through 2027. At least $1 trillion."
What it commits to (deployment): Orders and demand for Blackwell and Rubin chips through 2027 will be at least $1 trillion. He added that "we are going to be short" because he is certain computing demand will be higher.
Due: 2027 ("through 2027".)
Today (2026-10-17): not yet due. 2027 has not been reached.
Restates: huang-2025-10-28-gtc-dc-half-a-trillion-dollars-blackwell-rubin-through-2026
Notes: This raises the earlier $500 billion figure and extends it by a year. The words "Blackwell and Rubin" appear just before the quoted sentence.

huang-2026-03-23-lex-fridman-we-have-achieved-agi

Said: 2026-03-23 (Publication date on the transcript page (2026-03-23); the recording date was not shown.)
Occasion: Lex Fridman Podcast #494, "Jensen Huang: NVIDIA - The $4 Trillion Company & the AI Revolution"
Source (transcript): https://lexfridman.com/jensen-huang-transcript, at Section "AGI timeline", timestamp 01:55:06 to 01:58:28
Checked how: Fetched the transcript page with curl and read the whole exchange. Question asked (Fridman, 01:55:06 to 01:56:31): taking "this ridiculous definition" of AGI as an AI system able to essentially do his job, namely "start, grow, and run a successful technology company" worth "more than a billion dollars", "Is this five, 10, 15, 20 years away?"
The words: "I think it’s now. I think we’ve achieved AGI."
What it commits to (capability): In answer to when an AI will be able to start, grow and run a technology company worth over a billion dollars, he said that has already happened. He then qualified it: the company need not last ("you didn't say forever"), a viral app that briefly had billions of users could count, and "the odds of 100,000 of those agents building NVIDIA is zero percent".
Due: 2026 (Present-tense claim that it is already true as of March 2026.)
Today (2026-10-17): unfalsifiable. It rests on the interviewer's loose definition, which Huang read as allowing a short-lived success. I found no report of an AI-run company worth over a billion dollars (I did not search for one specifically), and he himself says AI building a company like Nvidia is a zero percent chance.
Notes: The repository panel row calls this statement "secondhand"; it is now seen at the source. He repeated it in softer form on 2026-08-26 (next row) and, per Fox Business coverage of 2026-09-08 (not a primary source, X post not opened), wrote on X "AGI has arrived" after OpenAI launched GPT-6 Astra.

huang-2026-08-26-earnings-call-already-achieved-agi-for-many-tasks

Said: 2026-08-26 (Date of Nvidia's second-quarter fiscal 2027 results release and earnings call (the release states 2026-08-26).)
Occasion: Nvidia second-quarter fiscal 2027 earnings call, answer to an analyst (Ben Reitzes of Melius Research) about recursive self-improvement and AGI
Source (transcript): https://www.webull.com/news/15497781742806016, at Answer to the analyst question that starts "I wanted to ask you a question, Jensen, about demand in a different way"
Checked how: Fetched two third-party transcripts of the call (Webull and platformaeronaut.com) with curl; both carry the same sentence. I did not find the transcript on Nvidia's own investor-relations page.
The words: "in a lot of ways, and for many tasks, we could say that we've already achieved AGI."
What it commits to (capability): He said that for many tasks we could say AGI has already been achieved, then called AGI milestones "kind of senseless at this point" and said what matters is that AI is doing useful work and generating profitable tokens.
Due: 2026 (Present-tense claim, made in August 2026.)
Today (2026-10-17): unfalsifiable. "For many tasks, we could say" has no test and no stated threshold; he then says the milestones do not matter.
Restates: huang-2026-03-23-lex-fridman-we-have-achieved-agi
Notes: Softer than March 2026 ("I think we've achieved AGI"). The panel row's "latest 2026-03" is therefore out of date.

Shane Legg

legg-2009-12-07-human-level-agi-2025

Said: 2009-12-07 (Exact day. The post's own header reads 'Posted on 7 December, 2009 by Shane'.)
Occasion: Blog post 'Tick, tock, tick, tock… BING' on his personal blog, vetta.org
Source (own words): https://www.vetta.org/2009/12/tick-tock-tick-tock-bing/, at Last paragraph of the post, just above the comments
Checked how: Fetched the page with curl on 2026-10-17 and read the post text and its date line.
The words: "my prediction for the last 10 years has been for roughly human level AGI in the year 2025"
What it commits to (timing): He had for ten years expected roughly human-level artificial general intelligence (AGI) around 2025, his most likely year. In the same post he says his expected value is 2028 and his 90% credibility range runs from 2018 to 2036, and that sceptics will deny it has happened when it does.
Due: 2025 ('in the year 2025' is his stated most likely year (the mode of his distribution), not a deadline; his own range runs to 2036.)
Today (2026-10-17): missed. On 2025-12-11 Legg said on Google DeepMind's own podcast that AI is not yet at his bar for AGI ('an artificial agent that can at least do the kinds of cognitive things people can typically do'): he says of today's systems "they're not there yet". He still gives 50% for 2028. The claim was a most-likely year inside a wide spread, and he predicted sceptics would deny it had happened, so some will dispute 'missed'. Evidence: https://spoken.md/episode/the-arrival-of-agi-with-shane-legg-co-founder-of-deepmind-1000740871915
Flag: contested: 2025 was his most likely year inside a stated range to 2036
Notes: The 90% range (2018 to 2036) is a separate, still-open part of the same post. Other evidence for the status: the same episode on Google DeepMind's YouTube channel, https://www.youtube.com/watch?v=l3u_FAv33G0 (2025-12-11); Legg's remark at the 2026-09-16 DeepMind Institute launch that it is premature to declare AGI achieved is reported only through the Financial Times via aiweekly.co (coverage, not used as the evidence). Judgement call flagged for John: 'missed' rests on Legg's own bar and his own later statement.

legg-2011-12-31-proto-agi-within-eight-years

Said: 2011-12-31 (Exact day. The post header reads 'Posted on 31 December, 2011 by Shane'.)
Occasion: Year-end blog post 'Goodbye 2011, hello 2012' on vetta.org
Source (own words): https://www.vetta.org/2011/12/goodbye-2011/, at First paragraph of the post; the 'acid test' is in his reply to Carl Shulman in the comments of 2011-12-31
Checked how: Fetched the page with curl on 2026-10-17 and read the post text, date line and his comment replies.
The words: "I expect to see an impressive proto-AGI within the next 8 years."
What it commits to (capability): By about the end of 2019, 'a system with basic vision, basic sound processing, basic movement control, and basic language abilities, with all of these things being essentially learnt rather than preprogrammed', able to 'solve a range of simple problems, including novel ones'. In the same post he leaves human-level AGI at a mean of 2028 and a mode of 2025 'under the assumption that nothing crazy happens like a nuclear war'.
Due: 2019 ('within the next 8 years' counted from the post date of 2011-12-31, so by 2019-12-31.)
Today (2026-10-17): unfalsifiable. Legg himself says a list of specific capabilities 'would somewhat miss the point'. His test is whether a smart person with an AI background, shown how the system works and then using it, 'would be highly likely to conclude that full human level AGI wasn't more than a decade away'. That is a judgement by a hypothetical observer, not a measurable bar.
Restates: legg-2009-12-07-human-level-agi-2025
Notes: Judgement call flagged for John. I found no source that describes the state of the field at end-2019 against this description, so I did not mark it missed. A reader who counts a system as 'basic and learnt' only if it handles all four senses together could mark it missed (no such system existed by 2019 to my knowledge, but I have no citation). Legg also writes in the comments that this 'might be my last annual prediction'.

legg-2023-10-26-agi-by-2028

Said: 2023-10-26 (Publication date of the episode page ('Oct 26, 2023'); the recording date is not stated on the page.)
Occasion: Dwarkesh Podcast interview with Dwarkesh Patel
Source (transcript): https://www.dwarkesh.com/p/shane-legg, at Transcript, Legg's answer starting at 0:34:38, before Patel's question at 0:37:13
Checked how: Fetched the episode page with curl on 2026-10-17 and read the printed transcript.
The words: "I think there's a 50% chance that we have AGI by 2028."
What it commits to (timing): A 50% chance of AGI by 2028. He adds: 'I'm not going to be surprised if it doesn't happen by then. You often hit unexpected problems in research and science and sometimes things take longer than you expect.'
Due: 2028 ('by 2028', stated as a 50% chance, not a certainty.)
Today (2026-10-17): not yet due. 2028 has not arrived. He gave a probability of one in two, so even a miss would not by itself show the estimate was wrong.
Restates: legg-2009-12-07-human-level-agi-2025
Notes: In the same interview the host reads back Legg's 2011 blog post (mean 2028, mode 2025). Legg's own definition in this period: human-level, later renamed 'minimal AGI'.

legg-2025-12-11-minimal-agi-2028

Said: 2025-12-11 (Release date of the episode (YouTube upload date 2025-12-11, same as the podcast listing).)
Occasion: Google DeepMind: The Podcast with Hannah Fry, episode 'The arrival of AGI'
Source (transcript): https://www.youtube.com/watch?v=l3u_FAv33G0, at Closing exchange, about 0:49:40 into the video
Checked how: Downloaded the English captions of the official Google DeepMind video with yt-dlp on 2026-10-17 and read the exchange. The captions are automatic, so punctuation is the caption's.
The words: "Hannah Fry: are you still 50/50 by 2028? Shane Legg: Yes,"
What it commits to (timing): Confirms a 50% chance of 'minimal AGI' by 2028 (Fry asks 'that's minimal AGI?' and he answers 'Yes'). For full AGI he says 'some years later. Could be three, four, five, six years later', and 'I think it'll be within a decade'.
Due: 2028 ('50/50 by 2028'; full AGI three to six years after that (2031 to 2034).)
Today (2026-10-17): not yet due. 2028 has not arrived.
Restates: legg-2023-10-26-agi-by-2028
Notes: Only Fry's side carries the date; Legg's words are 'Yes'. Earlier in the episode he defines minimal AGI as 'an artificial agent that can at least do the kinds of cognitive things people can typically do'. The Decoder reported this on 2025-12-14 (coverage, not used for the quote).

Demis Hassabis

hassabis-2025-04-20-agi-five-to-ten-years

Said: 2025-04-20 (Broadcast date. The CBS page says it is 'an updated version of a story first published on April 20, 2025' (page updated 2025-08-03).)
Occasion: 60 Minutes interview with Scott Pelley
Source (transcript): https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/, at Passage after the Gemini and Astra segment: Pelley asks 'On track for AGI in--'
Checked how: Fetched the CBS transcript page with curl on 2026-10-17 and read the exchange.
The words: "In the next five to ten years, I think."
What it commits to (timing): Answering Pelley's 'On track for AGI in--', he says AGI (an AI with the versatility of a human mind) is coming in the next five to ten years, with the hedge 'I think'.
Due: 2035 ('the next five to ten years' from 2025, so 2030 to 2035; 2035 is the outer end.)
Today (2026-10-17): not yet due. Neither end of the range has arrived.
Notes: The transcript page was updated 2025-08-03, so wording could differ slightly from the 2025-04-20 broadcast; I did not compare against the video.

hassabis-2025-04-20-end-of-disease-within-decade

Said: 2025-04-20 (Broadcast date (see first claim from this source).)
Occasion: 60 Minutes interview with Scott Pelley
Source (transcript): https://www.cbsnews.com/news/artificial-intelligence-google-deepmind-ceo-demis-hassabis-60-minutes-transcript/, at Drug-development passage; Pelley asks 'The end of disease?'
Checked how: Fetched the CBS transcript page with curl on 2026-10-17 and read the exchange.
The words: "I think that's within reach. Maybe within the next decade or so, I don't see why not."
What it commits to (capability): Ending disease with the help of AI is 'within reach', 'maybe within the next decade or so'. Just before, he says 'one day maybe we can cure all disease with the help of AI'.
Due: 2035 ('within the next decade or so' from 2025.)
Today (2026-10-17): not yet due. 2035 has not arrived. Heavily hedged ('maybe', 'or so', 'I don't see why not').
Notes: Judging 'end of disease' will need an agreed test; John may prefer to rule this one unfalsifiable as worded.

hassabis-2026-01-20-same-timeline-end-of-decade

Said: 2026-01-20 (Date of the session at the World Economic Forum Annual Meeting in Davos (YouTube upload date 2026-01-20).)
Occasion: 'The Day After AGI' panel with Dario Amodei, moderated by Zanny Minton Beddoes, World Economic Forum Annual Meeting 2026
Source (transcript): https://www.youtube.com/watch?v=NnVW9epLlTM, at Early in the session, about 0:03:16
Checked how: Downloaded the English captions of the official World Economic Forum video with yt-dlp on 2026-10-17 and read the exchange. Captions are automatic.
The words: "I think I'm still on the same kind of timeline."
What it commits to (timing): Asked by the moderator: 'you said a 50% chance of a system that can exhibit all the cognitive capabilities humans can by the end of the decade ... do you stand by your prediction', he says he is on the same timeline, and adds that natural-science work is harder to automate and that 'there may be one or two missing ingredients'.
Due: 2030 ('by the end of the decade' as put by the moderator and accepted by Hassabis; read as 2030 (could mean end of 2029).)
Today (2026-10-17): not yet due. The decade has not ended. A 50% chance, not a certainty.
Restates: hassabis-2025-04-20-agi-five-to-ten-years
Notes: The 50% and 'end of the decade' wording is the moderator's description of Hassabis's earlier statement; I did not locate his original 2025 words for it. His reply confirms it.

hassabis-2026-04-20-agi-2030

Said: 2026-04-20 (Date of Sequoia's AI Ascent 2026 event per Sequoia's own page (published 2026-05-08). The video was posted 2026-04-29. Sequoia's page lists Hassabis among attendees but does not say which day his talk was; I assumed the event day.)
Occasion: Conversation with Konstantine Buhler at Sequoia Capital's AI Ascent 2026
Source (transcript): https://www.youtube.com/watch?v=AFpeWo1GTeg, at Closing rapid-fire questions, about 0:25:25
Checked how: Downloaded the English captions of Sequoia's video with yt-dlp on 2026-10-17 and read the exchange. Captions are automatic. Event date from https://sequoiacap.com/article/ai-ascent-2026
The words: "No, 2030. I've been pretty consistent about that."
What it commits to (timing): Asked for his over-under year for AGI, he says 2030 and that he has been consistent about it. The video's own description says he explains 'why he believes AGI is achievable by 2030'.
Due: 2030 ('2030', given as a single year in a rapid-fire answer, with no stated probability in that exchange.)
Today (2026-10-17): not yet due. 2030 has not arrived.
Restates: hassabis-2025-04-20-agi-five-to-ten-years
Notes: The caption of the question reads 'Over under on distribution year of AGI', which is probably 'over-under on the date of AGI'. The definition of AGI used in this exchange is not stated.

Mustafa Suleyman

suleyman-2023-07-14-modern-turing-test-two-years

Said: 2023-07-14 (Exact day printed on the article ('July 14, 2023').)
Occasion: Essay by Suleyman in MIT Technology Review proposing a 'Modern Turing Test'
Source (own words): https://www.technologyreview.com/2023/07/14/1076296/mustafa-suleyman-my-new-turing-test-would-see-if-ai-can-make-1-million/, at Paragraph after the description of the test, beginning 'Something like this could be...'
Checked how: Fetched the article with curl on 2026-10-17 and read the text.
The words: "Something like this could be as little as two years away."
What it commits to (capability): An AI able to 'Go make $1 million on a retail web platform in a few months with just a $100,000 investment', researching and designing products, dealing with manufacturers, negotiating contracts and running marketing with minimal oversight (a human would still approve points, open a bank account and sign), 'could be as little as two years away'.
Due: 2025 ('as little as two years away' from 2023-07-14: the earliest he says it could happen (mid-2025). It is a lower bound, not a deadline.)
Today (2026-10-17): unfalsifiable. 'Could be as little as' names the earliest date it might happen, so missing 2025 does not contradict it. No report that any AI has passed the test has been found as of 2026-10-17, and in February 2026 Suleyman himself said it would happen 'this year'.
Flag: a lower bound ("as little as"), so missing the date proves nothing
Notes: Judgement call flagged for John: a reader who treats 'as little as two years' as a forecast of about 2025 could mark it missed. The test is restated in his 2023 book 'The Coming Wave' (not opened) and on X on 2026-01-05 (https://x.com/mustafasuleyman/status/2008208870204948746, coverage via Benzinga; not opened directly).

suleyman-2024-12-09-agi-five-to-seven-years

Said: 2024-12-09 (Publication date in the page's metadata (datePublished 2024-12-09); the interview was recorded earlier, date not stated.)
Occasion: Decoder podcast with Nilay Patel, printed transcript on The Verge
Source (transcript): https://www.theverge.com/24314821/microsoft-ai-ceo-mustafa-suleyman-google-deepmind-openai-inflection-agi-decoder-podcast, at Opening question about Sam Altman's Reddit remark that AGI can be reached on current hardware
Checked how: Fetched the transcript page with curl on 2026-10-17 and read the exchange; date from the page metadata.
The words: "I do think it is going to be plausible at some point in the next two to five generations."
What it commits to (timing): AGI cannot be done on current chips ('I don't think it can be done on [Nvidia] GB200s') but is plausible within two to five chip generations. He says 'I think within the next five to seven years since each generation takes 18 to 24 months now. So, five generations could be up to 10 years away depending on how things go.'
Due: 2031 ('within the next five to seven years' from late 2024 (2029 to 2031); he also names up to 10 years (2034) as an outer figure.)
Today (2026-10-17): not yet due. Neither end of the range has arrived.
Notes: Hedged: 'plausible', not 'will'. No '20 years out' claim by Suleyman was found.

suleyman-2026-02-12-modern-turing-test-this-year

Said: 2026-02-12 (YouTube upload date of the Financial Times video; the interview date itself is not stated.)
Occasion: Financial Times interview with editor Roula Khalaf, 'Mustafa Suleyman sets out Microsoft AI's goal of humanist superintelligence'
Source (transcript): https://www.youtube.com/watch?v=YTrBz6Z5c0E, at About 0:16:41 into the 21-minute video
Checked how: Downloaded the English captions of the official FT video with yt-dlp on 2026-10-17 and read the passage. Captions are automatic.
The words: "this year I think there are going to be models that can, you know, achieve that artificial capable intelligence and satisfy the modern Turing test."
What it commits to (capability): In 2026, models will reach what he calls artificial capable intelligence and pass the Modern Turing Test: an AI taking $100,000, inventing a product, building and marketing a business, and making a million dollars in a short period.
Due: 2026 ('this year', said in February 2026.)
Today (2026-10-17): not yet due. 2026 has not ended (2.5 months remain) and I found no report that any AI has done it.
Restates: suleyman-2023-07-14-modern-turing-test-two-years
Notes: No evidence of a pass found in searches on 2026-10-17; absence of reports is not proof. He says 'I think', so it is hedged.

suleyman-2026-02-12-white-collar-tasks-12-to-18-months

Said: 2026-02-12 (YouTube upload date of the Financial Times video; the interview date itself is not stated.)
Occasion: Financial Times interview with editor Roula Khalaf
Source (transcript): https://www.youtube.com/watch?v=YTrBz6Z5c0E, at About 0:17:10 into the 21-minute video
Checked how: Downloaded the English captions of the official FT video with yt-dlp on 2026-10-17 and read the passage. Captions are automatic.
The words: "most of those tasks will be fully automated by an AI within the next 12 to 18 months."
What it commits to (deployment): 'I think that we're going to have a human-level performance on most, if not all, professional tasks.' White-collar work done sitting at a computer (he names lawyer, accountant, project manager, marketing person): most of those tasks will be fully automated within 12 to 18 months.
Due: 2027 (12 to 18 months from February 2026, so February to August 2027.)
Today (2026-10-17): not yet due. The window has not closed (earliest end February 2027).
Notes: Wording is about tasks, not jobs. 'Fully automated' and 'most' will need a stated test. FT coverage of the same interview was widely reported around 2026-02-13.

Ilya Sutskever

sutskever-2023-07-05-superintelligence-this-decade

Said: 2023-07-05 (Exact day printed on the post ('July 5, 2023'; authors Jan Leike and Ilya Sutskever).)
Occasion: OpenAI blog post 'Introducing Superalignment', co-authored with Jan Leike
Source (own words): https://web.archive.org/web/2023/https://openai.com/blog/introducing-superalignment, at Second paragraph of the post
Checked how: Fetched the Wayback Machine copy with curl on 2026-10-17 (the live openai.com page returned 403 to curl) and read the text.
The words: "we believe it could arrive this decade."
What it commits to (timing): Superintelligence, which the post says 'seems far off now', 'could arrive this decade'. It is written in the first person plural for OpenAI, with Sutskever and Leike as named authors.
Due: 2029 ('this decade' written in 2023, read as the 2020s, so by the end of 2029 (2030 if counted as the decade ending 2030).)
Today (2026-10-17): not yet due. The decade has not ended. 'Could' makes it a possibility, not a firm forecast.
Notes: Joint corporate authorship: the words are Sutskever's only as co-author. Live page (canonical): https://openai.com/index/introducing-superalignment/

sutskever-2023-07-05-alignment-solved-in-four-years

Said: 2023-07-05 (Exact day printed on the post.)
Occasion: OpenAI blog post 'Introducing Superalignment', co-authored with Jan Leike
Source (own words): https://web.archive.org/web/2023/https://openai.com/blog/introducing-superalignment, at Section 'The new team', first paragraph (also the post's subtitle: 'To solve this problem within four years')
Checked how: Fetched the Wayback Machine copy with curl on 2026-10-17 and read the text.
The words: "Our goal is to solve the core technical challenges of superintelligence alignment in four years."
What it commits to (capability): The team he co-led would solve the core technical problems of making much-smarter-than-human AI follow human intent within four years. The post adds: 'While this is an incredibly ambitious goal and we're not guaranteed to succeed'.
Due: 2027 ('in four years' from 2023-07-05, so by mid-2027.)
Today (2026-10-17): not yet due. The deadline (July 2027) has not arrived. But OpenAI dissolved the Superalignment team in May 2024 after Sutskever and co-lead Jan Leike left, so the effort as described has ended; I found no statement from Sutskever withdrawing the goal. Evidence: https://www.popsci.com/technology/openai-dissolved-its-team-dedicated-to-preventing-rogue-ai/
Notes: The evidence URL is news coverage of the team's dissolution (May 2024), given only for context. John may prefer to treat this as withdrawn by events; the rule here requires the speaker to say so, and I found no such statement.

sutskever-2025-11-25-human-like-learning-five-to-twenty-years

Said: 2025-11-25 (Publication date of the episode page ('Nov 25, 2025'); the recording date is not stated.)
Occasion: Dwarkesh Podcast interview with Dwarkesh Patel
Source (transcript): https://www.dwarkesh.com/p/ilya-sutskever-2, at Transcript, 01:22:14 to 01:22:30
Checked how: Fetched the episode page with curl on 2026-10-17 and read the printed transcript.
The words: "I think like 5 to 20."
What it commits to (timing): Asked for his forecast for 'this system you're describing, which can learn as well as a human and subsequently, as a result, become superhuman', he answers 5 to 20 and confirms 'years' when the host repeats it.
Due: 2045 (5 to 20 years from late 2025, so 2030 to 2045; the later end is the point by which the claim is settled.)
Today (2026-10-17): not yet due. Neither end of the range has arrived.
Notes: The question quoted above is the host's, not Sutskever's. In the same interview he says stalling out of the current approach 'will all look very similar among all the different companies'.

Geoffrey Hinton

hinton-2016-10-27-stop-training-radiologists

Said: 2016-10-27 (Date of the conference per the organiser's own blog ('on October 27th, 2016'). The video was posted 2016-11-24.)
Occasion: Creative Destruction Lab conference 'Machine Learning and the Market for Intelligence', Rotman School, Toronto (panel answer)
Source (transcript): https://www.youtube.com/watch?v=2HMPRXstSvQ, at About 0:27 into the 1-minute clip 'Geoff Hinton: On Radiology' on the Creative Destruction Lab channel
Checked how: Downloaded the English captions of the organiser's video with yt-dlp on 2026-10-17 and read them. Event date from https://creativedestructionlab.com/blog/overview-of-the-mkt4intel/ ; the clip description names the conference.
The words: "people should stop training Radiologists now it's just completely obvious that within 5 years um deep learning is going to do better than Radiologists"
What it commits to (capability): Within five years deep learning would do better than radiologists, so radiologists should stop being trained. Right after, he adds 'it might be 10 years but we got plenty of Radiologists already'.
Due: 2021 ('within 5 years'; his fallback 'it might be 10 years' falls on 2026-10-27, ten days after the date of this check.)
Today (2026-10-17): missed. Five years on (2021) radiologists had not been replaced, and demand has kept rising: the American College of Radiology reported on 2026-02-05 that workforce shortage is the field's biggest threat and that the workforce is projected to be larger by 2055. Hinton himself later said he had spoken too broadly (reported by the New York Times; I could not open it). Evidence: https://prod.acr.org/Clinical-Resources/Publications-and-Research/ACR-Bulletin/2026/radiologist-shortage-work-force-update
Flag: contested: "do better than" has been met in narrow studies; the practical outcome (stop training them) is what failed
Notes: Judgement call flagged for John: the claim is that deep learning will 'do better than' radiologists, which AI has done in some narrow image-reading studies; the part that clearly failed is the practical implication (stop training them). Hinton's partial walk-back is by coverage only (Fortune 2026-07-19, https://fortune.com/article/ai-godfather-radiologists-obsolete-salaries-up-to-571k-demand-growing/ ; it reports US active radiologists up about 10% over ten years). Not 'withdrawn': he said the direction was right. Not in the repository's statements file.

hinton-2023-05-03-five-to-twenty-years

Said: 2023-05-03 (Exact day and time from the post's own timestamp: 2023-05-03 09:08 UTC.)
Occasion: Reply on Twitter (now X) from his own account
Source (own words): https://twitter.com/geoffreyhinton/status/1653687894534504451, at The reply itself (a reply to @dontfearai)
Checked how: Read the post through Twitter's public embed endpoint (cdn.syndication.twimg.com, tweet id 1653687894534504451) on 2026-10-17: author geoffreyhinton, created_at 2023-05-03T09:08:12Z.
The words: "I now predict 5 to 20 years but without much confidence."
Coded statement row: hinton-2023-05
What it commits to (timing): When AI will be smarter than a human: 5 to 20 years, 'without much confidence'. He continues: 'It's possible that I am totally wrong about digital intelligence overtaking us. Nobody really knows which is why we should worry now.'
Due: 2043 (5 to 20 years from 2023, so 2028 to 2043.)
Today (2026-10-17): not yet due. Neither end of the range has arrived.
Notes: The reply's text does not say what is being predicted; the question it answers is quoted in Kaj Sotala's collection (https://www.greaterwrong.com/posts/bLvc7XkSSnoqSukgy/a-brief-collection-of-hinton-s-recent-comments-on-agi-risk) as asking when AI would be smarter than a human, after his earlier 30 to 50 years.

hinton-2023-10-08-reason-better-than-us-in-five-years

Said: 2023-10-08 (Broadcast date; the CBS page says it is 'an updated version of a story first published on Oct. 8, 2023'.)
Occasion: 60 Minutes interview with Scott Pelley
Source (transcript): https://www.cbsnews.com/news/geoffrey-hinton-ai-dangers-60-minutes-transcript/, at Early in the transcript, after the exchange on whether ChatGPT-4 'understands'
Checked how: Fetched the CBS transcript page with curl on 2026-10-17 and read the exchange.
The words: "I think in five years' time it may well be able to reason better than us."
Coded statement row: hinton-2023-07
What it commits to (capability): Answering Pelley's 'And in five years' time?', he says AI (he had just said GPT-4 'definitely understands') may well reason better than people.
Due: 2028 ('in five years' time' from 2023.)
Today (2026-10-17): not yet due. 2028 has not arrived. 'May well' is hedged, and 'reason better than us' has no agreed test; some will say today's reasoning models already do on some tasks.
Notes: Not marked hit because of the hedge and the lack of a test.

hinton-2024-12-06-nobel-interview-five-to-twenty

Said: 2024-12-06 (The Nobel Prize page says the interview was 'recorded on 6 December 2024 during Nobel Week in Stockholm'.)
Occasion: Official Nobel Prize interview with the 2024 physics laureate
Source (transcript): https://www.nobelprize.org/prizes/physics/2024/hinton/1925103-interview-transcript/, at Answer to 'How much time do we have before AI outsmarts us?'
Checked how: Fetched the transcript page with curl on 2026-10-17 and read the answer.
The words: "My guess is in between five and 20 years from now, there's a good chance, a 50% chance, we'll get AI smarter than us."
Coded statement row: hinton-2024-02
What it commits to (timing): A 50% chance of AI smarter than us in five to twenty years. He then says: 'Actually, that was my guess a year ago. I guess my guess now is between four and 19 years.'
Due: 2043 ('between four and 19 years' from 2024, so 2028 to 2043.)
Today (2026-10-17): not yet due. Neither end of the range has arrived.
Restates: hinton-2023-05-03-five-to-twenty-years
Notes: He also says 'It may be much longer, it's just possible it's a bit shorter' and 'I think it's quite likely to have happened in 20 years time'.

Yann LeCun

lecun-2024-10-16-human-level-ai-several-years-if-not-a-decade

Said: 2024-10-16 (Exact day and time from the post's timestamp: 2024-10-16 15:31 UTC.)
Occasion: Post on X from his own account, quoting another user's post asking why he disagrees with Sam Altman and Dario Amodei
Source (own words): https://x.com/ylecun/status/1846574605894340950, at First line of the post
Checked how: Read the full post text through Twitter's public embed endpoint and a third-party mirror API (api.fxtwitter.com) on 2026-10-17; author ylecun, created 2024-10-16T15:31:08Z. x.com itself was not opened.
The words: "I said that reaching Human-Level AI "will take several years if not a decade.""
Coded statement row: lecun-2024-07
What it commits to (timing): Reaching human-level AI 'will take several years if not a decade'. He then reads Sam Altman's 'several thousand days' as 6 to 9 years ('So we're not in disagreement') and adds 'the distribution has a long tail: it could take much longer than that. In AI, it almost always takes longer.'
Due: 2034 ('a decade' from 2024; he also allows a long tail beyond that.)
Today (2026-10-17): not yet due. 2034 has not arrived. He gave a long-tailed distribution, so a miss would not settle it.
Notes: He quotes his own earlier words ('I said'), so the original statement is older than this post; I did not find the earliest instance.

lecun-2024-10-16-not-in-the-next-year-or-two

Said: 2024-10-16 (Exact day and time from the post's timestamp: 2024-10-16 15:31 UTC.)
Occasion: Post on X from his own account
Source (own words): https://x.com/ylecun/status/1846574605894340950, at Last line of the post
Checked how: Read the full post text through a third-party mirror API (api.fxtwitter.com) on 2026-10-17; the first public embed endpoint cut the post off before this line.
The words: "In any case, it's not going to be in the next year or two."
Coded statement row: lecun-2024-07
What it commits to (timing): Human-level AI will not arrive within a year or two of 2024-10-16, that is not before about October 2026.
Due: 2026 ('the next year or two' from 2024-10-16, so to about October 2026.)
Today (2026-10-17): not yet due. The two years ended on about 2026-10-16 but the due year, 2026, has not finished. On the evidence I found it is on course to hold: Legg said on 2025-12-11 current AI is 'not there yet', and Hassabis on 2026-01-20 said 'one or two missing ingredients' remain. Against that, executives at Nvidia and OpenAI have claimed in 2026 that AGI has arrived (reported via the Financial Times, coverage only), so what counts as 'human-level' is contested.
Notes: Likely to be a hit if no one's claim of AGI is accepted by year end. Definition of 'human-level AI' is LeCun's, not stated in the post.

lecun-2025-01-23-llms-obsolete-within-five-years

Said: 2025-01-23 (Session date at the World Economic Forum Annual Meeting 2025 in Davos (the YouTube upload date is also 2025-01-23; TechCrunch reported the session the same day).)
Occasion: 'Debating Technology' town hall with Dava Newman, moderated by Ina Fried, World Economic Forum Annual Meeting 2025 (answer to an audience question)
Source (transcript): https://www.youtube.com/watch?v=MohMBV3cTbg, at Answer to an audience question about where AI will be in five years, about 0:26:12
Checked how: Downloaded the English captions of the official World Economic Forum video with yt-dlp on 2026-10-17 and read the answer. Captions are automatic and have no speaker labels; I attribute the answer to LeCun from the content ('my colleagues and I', Meta's research) and the session's speaker list (Newman, LeCun, Fried).
The words: "within five years nobody in the right mind would use them anymore at least not as kind of the central component of an AI system"
What it commits to (capability): Large language models: 'the shelf life of the current Paradigm ... is fairly short probably 3 to five years'. Within five years nobody would use them as the central component of an AI system.
Due: 2030 ('within five years' from 2025-01-23.)
Today (2026-10-17): not yet due. 2030 has not arrived.
Notes: 'Them' refers to large language models (the caption says 'large language model' in the previous sentence). Reworded in print as 'nobody in their right mind will use autoregressive LLMs' in later coverage; I did not use the later wording. Not in the repository's statements file as a separate row.

lecun-2025-01-23-new-paradigm-in-three-to-five-years

Said: 2025-01-23 (Session date at the World Economic Forum Annual Meeting 2025 (see previous claim).)
Occasion: 'Debating Technology' town hall, same answer as the previous claim
Source (transcript): https://www.youtube.com/watch?v=MohMBV3cTbg, at Same answer, a little later, about 0:26:56
Checked how: Downloaded the English captions of the official World Economic Forum video with yt-dlp on 2026-10-17. Same speaker-attribution caveat as the previous claim.
The words: "within 3 to 5 years we'll have system that are complete different Paradigm they may have some level of Common Sense"
Coded statement row: lecun-2025-02
What it commits to (capability): Conditional: 'if the plan that we're working on succeeds ... with the timetable that we hope', within 3 to 5 years there will be systems built on a completely different paradigm that 'may have some level of Common Sense' and may learn how the world works from observing it.
Due: 2030 ('within 3 to 5 years' from 2025-01-23, so 2028 to 2030.)
Today (2026-10-17): not yet due. Neither end of the range has arrived. It is conditional on his own team's plan working.
Notes: This is not a date for human-level AI; the repository row notes the same. A mirror of this session is TechCrunch 2025-01-23 (coverage).

Kai-Fu Lee

kaifulee-2017-04-27-half-of-jobs-next-decade

Said: 2017-04-27 (Publication date printed on CNBC's article ('Published Thu, Apr 27 2017'). The interview was at the Global Mobile Internet Conference in Beijing; its exact day is not stated.)
Occasion: CNBC interview at the Global Mobile Internet Conference, Beijing (CNBC article by Sophia Yan)
Source (coverage only): https://www.cnbc.com/2017/04/27/kai-fu-lee-robots-will-replace-half-of-all-jobs.html, at First paragraphs of the article
Checked how: Fetched the article with curl on 2026-10-17. His words appear only as CNBC quotes them; the CNBC video page loaded but its video could not be extracted, so I did not see the recording.
The words: none: coverage only
What it commits to (deployment): As CNBC reports it: robots and AI will 'probably' replace about 50 percent of human jobs, and he expected that displacement 'to occur in the next 10 years'.
Due: 2027 ('the next 10 years' from 2017 as reported by CNBC (paraphrase, not his quoted words).)
Today (2026-10-17): not yet due. 2027 has not arrived.
Notes: To upgrade to a quote, the CNBC video or a transcript is needed. Later he gave a different figure and horizon: 40 to 50 percent of US jobs in 15 years (CNBC, 2018-12-14, not opened) and 40 percent of the world's jobs in 15 years (see the 2019 claim).

kaifulee-2019-01-13-forty-percent-of-jobs-fifteen-years

Said: 2019-01-13 (Air date of the 60 Minutes segment ('Sunday, Jan. 13', per CBS's advance release reposted by WDEF on 2019-01-09). I saw the segment only in a 2023-12-30 compilation upload that describes it as 'From January 2019'.)
Occasion: 60 Minutes interview with Scott Pelley, 'The Oracle of AI'
Source (transcript): https://www.youtube.com/watch?v=aZ5EsdnpLMI, at About 0:10:16 into the compilation video (the Kai-Fu Lee segment is its first story)
Checked how: Downloaded the English captions of the official 60 Minutes upload with yt-dlp on 2026-10-17 and read the segment; air date from https://www.wdef.com/?p=112844 (CBS release). Captions are automatic.
The words: "all together in 15 years that's going to uh displace uh about 40% of jobs in the world"
What it commits to (deployment): Automation will displace about 40 percent of the world's jobs in about 15 years. He then says '40% of jobs in the world will be displaced by technology I would say displaceable', and later 'whether it's 15 or 25 years is coming faster than the previous revolutions'. Drivers' jobs 'will be disrupted more in the 15 to 20 year time frame'.
Due: 2034 ('in 15 years' from 2019; he also says 'whether it's 15 or 25 years', which would be 2044.)
Today (2026-10-17): not yet due. 2034 has not arrived. He softens it to 'displaceable' jobs, not jobs that will disappear.
Notes: Other sources date the same claim to 2033 (from his 2018 book 'AI Superpowers' and a 2018 interview); I did not open those. His quote contains spoken fillers ('uh') as in the caption.

kaifulee-2019-01-13-agi-not-within-thirty-years

Said: 2019-01-13 (Air date of the 60 Minutes segment (see previous claim).)
Occasion: 60 Minutes interview with Scott Pelley, 'The Oracle of AI'
Source (transcript): https://www.youtube.com/watch?v=aZ5EsdnpLMI, at About 0:12:02 into the compilation video
Checked how: Downloaded the English captions of the official 60 Minutes upload with yt-dlp on 2026-10-17 and read the segment. Captions are automatic.
The words: "if you're talking about AGI artificial general intelligence I would say not within the next 30 Years and possibly never"
What it commits to (timing): AGI will not arrive within the next 30 years from 2019, 'and possibly never'.
Due: 2049 ('not within the next 30 years' from 2019 (a claim that nothing happens before 2049).)
Today (2026-10-17): not yet due. 2049 has not arrived. In September 2026 he said the opposite in substance: AI is 'at the footsteps of AGI' and about two years from his bar for AGI (see the 2026 claim). I found no statement from him that the 2019 view was wrong, so I did not mark it withdrawn. Evidence: https://www.youtube.com/watch?v=c4w3njBhYVk
Notes: Judgement call flagged for John: this may count as withdrawn in substance. The rule here (speaker says it no longer holds) is not met by anything I found.

kaifulee-2026-09-23-agi-about-two-years

Said: 2026-09-23 (YouTube upload date. The recording date is not stated; he refers to his book as coming out 'mid-September', so it may have been recorded earlier.)
Occasion: Endgame #279 podcast with Gita Wirjawan
Source (transcript): https://www.youtube.com/watch?v=c4w3njBhYVk, at Near the end, about 1:40:38 into the 1 hour 49 minute video
Checked how: Downloaded the English captions of the Endgame channel's video with yt-dlp on 2026-10-17 and read the passage. Captions are automatic.
The words: "But answering questions and doing things are quite different. I think we're probably still maybe two years away from that."
What it commits to (timing): Using his definition of AGI ('for cognitive tasks, when AI can do 90% of the human tasks better than 90% of the people'): AI can already answer 90% of questions better than 90% of people, but doing the tasks is 'probably still maybe two years away'. If AGI also includes physical robots, 'that adds another five to ten years'.
Due: 2028 ('maybe two years away' from September 2026 for cognitive tasks; 2033 to 2038 if embodied AI is included.)
Today (2026-10-17): not yet due. 2028 has not arrived. Hedged: 'probably', 'maybe'.
Notes: He also says he is '95% confident' AI's sentience will not be replicated 'in the next 100 years' and that an all-encompassing AGI 'is impossible to measure', so he 'wouldn't want to put a date by that at all'.

Rendered from docs/claims-ledger-seed-2026-10-17.md at version a7bd17a.