Yudkowsky's earlier analysis of AI foom explicitly did not rule out the current AI development scenario.
Unresolvable The public record could not settle it.
Factual: Statements about the present or the past, checked against the record: did they hold up?
The second judge said mostly true; the verdict shown is the first judge's.
The evidence is the person's or their organisation's own records.
What would count as true or false
True: Yudkowsky's earlier writings on AI takeoff explicitly allow for, or name as possible, a scenario like the one current in 2024. Such writings include 'Intelligence Explosion Microeconomics' (2013), the 2008 Hanson–Yudkowsky FOOM debate and related posts. The scenario is large deep-learning models with gradual, compute-driven progress before any recursive self-improvement. 'Explicitly' requires text that states this possibility, not just an absence of a contrary prediction. Mostly true: his past analysis clearly hedges to allow such scenarios, but the wording is general rather than describing them specifically. Mixed: the earlier writings contain both statements allowing it and statements that appear to predict otherwise. Mostly false: the scenario is not ruled out only implicitly, or the earlier writings mainly predicted a contrary scenario such as hand-coded AGI or a sudden local foom without large-scale compute. False: the earlier analysis explicitly ruled out or dismissed this scenario.
Set before the research began.
Evidence (4)
-
E1 MIRI MIRI (hosted at coefficientgiving.org) · 27 November 2025 Own record
“AI progress will be dominated for some time by the contributions of the world economy to AI research, rather than by any one AI’s internal self-improvement.”
-
E2 MIRI MIRI (hosted at coefficientgiving.org) · 27 November 2025 Own record
“A collective of interacting AIs will have k ≈ 1+ and grow at a manageable, human-like exponential pace, just like the world economy”
-
E3 MIRI MIRI (hosted at coefficientgiving.org) · 27 November 2025 Own record
“But of course (so far as the author believes) such other outcomes would be even harder to defend in depth.”
-
E4 MIRI MIRI (hosted at coefficientgiving.org) · 27 November 2025 Own record
“Even if it were a matter of hardware rather than mostly software, the threshold level of “required hardware” would be far more uncertain than Moore’s Law”
Reasoning
The Intelligence Explosion Microeconomics paper describes a scenario of AI progress driven by the world economy and of AIs growing at a human-like pace (E1, E2). The quoted passages do not show whether the author allows this scenario or disputes it, and E3 and E4 do not clearly predict otherwise. The items also do not establish the paper's date or that Yudkowsky wrote it, so the evidence does not settle the claim under the criteria.
Cite this claim
Eliezer Yudkowsky, “Yudkowsky's earlier analysis of AI foom explicitly did not rule out the current AI development scenario.” (said 8 March 2024). Verdict: unresolvable. Reliability API, claim c_59d372e5b821, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_59d372e5b821
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More factual claims by Eliezer Yudkowsky
-
Mixed
“I was observing that obvious-seeming hypothesis myself, sometime around the time of GPT 3.5.”
Yudkowsky himself observed the persona-selection-like hypothesis around the time of GPT 3.5.
-
Mostly true
“The test-passing preference stands on its own once formed.”
-
Mostly true
“LLMs surely have neural circuits they learned to model text streams from fictional and nonfictional personas that are not lying, deceiving, cheating.”
-
Mostly true
“The predict-as-if-honest circuitry is contextual, invoked to predict some streams of text but not others.”
-
Mostly true
“Presumably, somewhere along the way LLMs learn a dispreference for tests that don't pass, independently of how that gets prevented or whether the new plan does what the user really wanted.”
Verdicts are made automatically; how verdicts are made.