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 was expecting there to be more of an issue, more of a sense of like, different discoveries.”
Eliezer Yudkowsky, interviewed by Lex Fridman March 2023 "Dangers of AI and the End of Human Civilization" Yudkowsky on Lex Fridman (LessWrong)
Yudkowsky previously expected AI progress to involve more distinct discoveries, rather than scaling an existing architecture.
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.
The evidence is the person's or their organisation's own records.
What would count as true or false
Judged as of 2023-03; the claim concerns Yudkowsky's own earlier expectations. True: Yudkowsky's earlier writings (for example the Hanson–Yudkowsky FOOM debate of 2008, or posts and papers before 2020) show he expected AI progress to come mainly from multiple distinct conceptual discoveries or insights, rather than from scaling an existing architecture such as transformers. Mostly true: his earlier writings mainly stress insights over scale, with some minor acknowledgment of the role of compute or scaling. Mixed: his earlier writings are about evenly split between expecting distinct discoveries and expecting scale or compute to dominate. Mostly false: his earlier writings mostly predicted that scaling would be important, with only occasional emphasis on discoveries. False: his documented earlier views clearly predicted progress driven by scaling an existing architecture.
Evidence
- Intelligence Explosion Microeconomics Machine Intelligence Research Institute (author: Eliezer Yudkowsky) Own record
“I no longer think that Moore’s Law will play much of a role in the intelligence explosion, partially because I expect returns on algorithms to dominate, and partially because I would expect an AI to prefer ways to scale itself onto more existing hardware rather than waiting for a new generation of chips to be produced in Intel-style factories.”
- Intelligence Explosion Microeconomics Machine Intelligence Research Institute (author: Eliezer Yudkowsky) Own record
“I identify the key issue as returns on cognitive reinvestment—the ability to invest more computing power, faster computers, or improved cognitive algorithms to yield cognitive labor which produces larger brains, faster brains, or better mind designs.”
- Intelligence Explosion Microeconomics Machine Intelligence Research Institute (author: Eliezer Yudkowsky) Own record
“Although I used to go around asking, “If Moore’s Law says that computing speeds double every eighteen months, what happens when computers are doing the research?””
Reasoning
In his MIRI paper Intelligence Explosion Microeconomics, Yudkowsky writes that he expects "returns on algorithms to dominate" over hardware trends such as Moore's Law (E1). That fits an earlier expectation that progress would come from insights rather than scale. The same paper still lists more computing power and faster computers as channels alongside better algorithms (E2), and he says he once framed AI progress around Moore's Law (E3), so compute and scaling are acknowledged. The evidence is undated and does not directly contrast "distinct discoveries" with scaling a single architecture, so the claim fits the "mostly true" level of the criteria.
Cite this claim
Eliezer Yudkowsky, “Yudkowsky previously expected AI progress to involve more distinct discoveries, rather than scaling an existing architecture.” (said March 2023). Verdict: mostly true. Reliability API, claim c_e6850918928e, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_e6850918928e
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.