Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
76 70 82 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
60 42 78
All types combined: 74 (likely 68 to 80, high confidence), from 79 claims: 71 factual claims, 8 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Eliezer Yudkowsky's claims
46 claims
“I have refused to deploy timelines with fancy probabilities on them consistently for many years”
Eliezer Yudkowsky, interviewee 6 April 2023 Eliezer Yudkowsky - Why AI Will Kill Us, Aligning LLMs, Nature of Intelligence, SciFi, & Rationality (Dwarkesh Podcast)
Eliezer Yudkowsky has consistently refused for many years to publish AI timelines with detailed probabilities.
Mostly true True in substance, wrong in a detail.
The second judge would not resolve this claim; the verdict shown is the first judge's.
What would count as true or false
Judge as of 6 April 2023, looking at Yudkowsky's public record over the preceding years (roughly 2013 to 2023 or longer). True: his writings, posts and interviews in that period contain no AI timelines stated with specific numerical probabilities (for example 'X% chance of AGI by year Y'), and he is documented as declining to give them. Mostly true: he generally refused, but there are one or two minor or offhand numerical statements (for example a casual bet) that do not amount to a timeline forecast. Mixed: there are several instances of probabilistic timelines alongside a general pattern of refusal. Mostly false: he repeatedly gave probabilistic timelines in that period, although he sometimes declined. False: he regularly published detailed probabilistic AI timelines during those years.
Evidence
- Eliezer Yudkowsky LongTermWiki · 23 February 2026
“In a 2019 post, he stated he had "a policy for over a decade of not putting numbers on a few things, one of which is AGI timelines and one of which is non-relative doom probabilities," citing instability in his own estimates as the reason.”
- Eliezer Yudkowsky LongTermWiki · 23 February 2026
“In 2001, he reportedly believed his team could develop a "final stage AI" that would "reach transhumanity sometime between 2005 and 2020, probably around 2008 or 2010."”
- Timelines to Transformative AI: an investigation — LessWrong LessWrong (cross-post by Convergence Analysis author Zershaaneh Qureshi) · 26 March 2024 Own record
“he did note *in 2023* that the timeline for AGI "could be three years. It could be 15 years. We could get that AI winter I was hoping for, and it could be 16 years. I'm not really seeing 50 without some kind of giant civilizational catastrophe."”
- Biology-Inspired AGI Timelines: The Trick That Never Works — AI Alignment Forum AI Alignment Forum · 1 December 2021 Own record
“But *timing* the novel development correctly? *That* is almost never done, not until things are 2 years out, and often not even then.”
Reasoning
An independent secondary source (E1) reports that in a 2019 post Yudkowsky described a policy, held for over a decade, of not putting numbers on AGI timelines. His 2021 essay (E4) argues that correctly timing novel developments is almost never done until they are about two years out, and often not even then, which fits that refusal. The 2023 remark quoted in E3 gives rough year ranges with no probabilities attached, so it reads as an offhand comment rather than a probabilistic forecast. E2 gives dates from 2001, well before the period being judged. Because the record is secondhand and incomplete, and he did give offhand numbers, the verdict is mostly true rather than fully true.
Cite this claim
Eliezer Yudkowsky, “Eliezer Yudkowsky has consistently refused for many years to publish AI timelines with detailed probabilities.” (said 6 April 2023). Verdict: mostly true. Reliability API, claim c_280f3104de87, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_280f3104de87
Claims without a verdict
Is a verdict here wrong, or do you speak for Eliezer Yudkowsky? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.