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?
81 76 87 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
Insufficient record: 2 predictions resolved, 5 needed for a score.
-
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
65 45 83
All types combined: 80 (likely 75 to 85, high confidence), from 84 claims: 78 factual claims, 6 promises. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Demis Hassabis's claims
42 claims
“I think you do sometimes get surprising improvement in other domains when you improve in a specific domain.”
Demis Hassabis, CEO of Google DeepMind 28 February 2024 Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold (Dwarkesh Podcast)
Demis Hassabis thinks that improving in one specific domain sometimes produces surprising improvement in other domains; as an example, he said that when large models improve at coding, that can improve their general reasoning.
Mostly true True in substance, wrong in a detail.
What would count as true or false
The claim is about what Hassabis himself said and thinks. True: a recording or transcript of his February 2024 remarks shows him saying that improving in one specific domain sometimes brings surprising gains in other domains, and giving the example that better coding ability in large models can improve their general reasoning. Mostly true: he expressed the view, but the coding example is worded somewhat differently or is attributed loosely. Mixed: he expressed the general view but not the coding example, or vice versa. Mostly false: the remarks are substantially misrepresented, e.g. he expressed doubt rather than this view. False: the transcript shows he did not say this, or said the opposite.
Evidence
- Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold podscripts.co · 28 February 2024
“Well, I think you, first of all, I think you do actually sometimes get surprising improvement in other domains when you improve in a specific domain.”
- The Scaling Era: An Oral History of AI, 2019–2025 Stripe Press · 7 October 2025
“You do sometimes get surprising improvement in other domains.”
- Notes on Dwarkesh Patel’s Podcast with Demis Hassabis thezvi.wordpress.com (Zvi Mowshowitz) · 1 March 2024
“Demis says often it does, such as coding improving reasoning (which also happens in humans), and he expects more chain transfer.”
Reasoning
A third-party transcript of the February 28, 2024 Dwarkesh Podcast interview (E1) has Hassabis saying he thinks you do sometimes get surprising improvement in other domains when you improve in a specific domain. The Stripe Press oral history (E3) attributes the same view to him. The example that better coding improves reasoning is supported only by a commentator's paraphrase (E4), not by a verbatim transcript quote. Under the criteria, that fits "mostly true" (the coding example is attributed loosely) rather than "true".
Cite this claim
Demis Hassabis, “Demis Hassabis thinks that improving in one specific domain sometimes produces surprising improvement in other domains; as an example, he said that when large models improve at coding, that can improve their general reasoning.” (said 28 February 2024). Verdict: mostly true. Reliability API, claim c_5c72b810ecef, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_5c72b810ecef
Claims without a verdict
Is a verdict here wrong, or do you speak for Demis Hassabis? 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.