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
21 claims
“We learned that that's false, but we did not learn that you can have a thing that can redesign a tree and build an AI much smarter than itself, that builds an AI much smarter than itself, and that you survive that because that part did not happen.”
Eliezer Yudkowsky, interviewee September 2025 Interview with Eliezer Yudkowsky on Rationality and Systematic Misunderstanding of AI Alignment (LessWrong)
No AI has yet been shown to redesign a tree and build a much smarter AI that builds a still smarter AI; that scenario has not occurred.
Resolved true The record shows it happened or was so.
What would count as true or false
Judged as of September 2025. The claim is that no AI has yet autonomously designed and built a much smarter AI that in turn built a still smarter AI, i.e. a multi-step recursive self-improvement chain. TRUE: credible sources from around September 2025 (AI lab reports, technical literature, expert surveys) show no documented case of an AI system autonomously creating a substantially more capable successor that itself created a substantially more capable successor. MOSTLY TRUE: no multi-step chain had occurred, but there was a documented case of an AI system substantially and largely autonomously creating one much more capable successor, which the claim's wording partly glosses over. MIXED: there was a documented case close to the chain, e.g. two generations of significant AI-driven improvement, but disputed as to how autonomous it was or how much smarter the results were. MOSTLY FALSE: there was a documented case of the chain occurring, with only minor caveats about autonomy or the size of the gains. FALSE: there was a clearly documented case, by September 2025, of an AI building a much smarter AI that then built a still smarter AI.
Evidence
- AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms Google DeepMind · 14 May 2025
“By finding smarter ways to divide a large matrix multiplication operation into more manageable subproblems, it sped up this vital kernel in Gemini’s architecture by 23%, leading to a 1% reduction in Gemini's training time.”
- Can Large Language Models Actually Discover New Algorithms? IEEE Spectrum · 14 May 2025
“for example, AlphaEvolve’s optimization of Gemini training may be seen as the beginning of recursively self-improving AI, which some worry would lead to a runaway intelligence explosion referred to as the singularity. The DeepMind team maintains that that is not their goal, of course.”
- Why recursive self-improvement suddenly became a serious question | IBM IBM Think · 16 September 2026
“Researchers have not shown that those abilities can combine into an autonomous cycle in which each generation reliably creates a substantially more capable successor, they said.”
Reasoning
A documented case of AI contributing to AI training is AlphaEvolve (E1, May 2025), which sped up a kernel in Gemini's architecture by 23%, leading to a 1% reduction in Gemini's training time. It sped up a kernel in Gemini's architecture by 23%, which cut Gemini's training time by 1%. That is a small efficiency gain, not an AI autonomously building a much smarter successor. IEEE Spectrum (E2) says this may be seen as "the beginning" of recursively self-improving AI, while reporting that DeepMind says that is not its goal. IBM (E3, September 2026) reports that researchers had still not shown an autonomous cycle in which each generation creates a substantially more capable successor, which implies no such chain had occurred by September 2025.
Cite this claim
Eliezer Yudkowsky, “No AI has yet been shown to redesign a tree and build a much smarter AI that builds a still smarter AI; that scenario has not occurred.” (said September 2025). Verdict: resolved true. Reliability API, claim c_7cd0a5dab6f6, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_7cd0a5dab6f6
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.