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 69 82 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
56 42 70 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 4 promises resolved, 5 needed for a score.
All types combined: 72 (likely 66 to 78, high confidence), from 93 claims: 76 factual claims, 17 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Dario Amodei's claims
28 claims
“I'd guess the number today is maybe ~4x/year.”
Dario Amodei, essay author 13 October 2022 On DeepSeek and Export Controls (darioamodei.com)
The current rate of algorithmic curve shift is maybe about 4x per year.
Resolved true The record shows it happened or was so.
What would count as true or false
The claim is hedged ('maybe ~4x/year'); judge whether the rate at which algorithmic progress cut the compute needed to reach a given performance level was roughly 4x per year as of October 2022. True: credible estimates for language models or deep learning broadly around 2022 (e.g., Epoch AI's algorithmic progress studies) fall within about 2x–8x per year, so ~4x is consistent with them. Mostly true: estimates cluster a little outside that band but the order of magnitude matches (e.g., 1.5x–10x). Mixed: estimates vary so widely (e.g., from under 1.5x to over 10x per year) that ~4x is neither supported nor contradicted. Mostly false: the best estimates are clearly away from 4x (e.g., around 1.3x or 15x+ per year) but some real research supports a figure near it. False: credible estimates show algorithmic efficiency gains far from 4x per year (well under 1.5x or well over 15x).
Evidence
- Algorithmic progress in language models Epoch AI · 12 March 2024
“We find that the level of compute needed to achieve a given level of performance has halved roughly every 8 months, with a 95% confidence interval of 5 to 14 months.”
- ALGORITHMIC PROGRESS IN LANGUAGE MODELS Anson Ho, Tamay Besiroglu, Ege Erdil, et al. (Epoch AI authors), hosted at pdf.stafforini.com · 8 October 2026
“we find that the compute required to reach a set performance threshold has halved approximately every 8 months”
- A compute-based framework for thinking about the future of AI Epoch AI · 31 May 2023
“Erdil and Besiroglu 2022 estimated that the training compute required to reach a fixed level of performance on ImageNet has been cutting in half roughly every nine months, albeit with wide uncertainty over that value.”
- Algorithmic progress in language models Epoch AI · 12 March 2024
“Lastly, there is substantial uncertainty about the most appropriate way to model algorithmic progress, given that there are many possible approaches with different predictions”
Reasoning
Epoch AI's study of language models (E1, E2) found that the compute needed to reach a fixed performance level halved about every 8 months. That works out to roughly 2.8x per year, and its 95% confidence interval of 5 to 14 months spans about 1.8x to 5.3x per year. Epoch AI also cites a 2022 ImageNet estimate of halving about every 9 months, roughly 2.5x per year (E3). The authors say there is substantial uncertainty about how best to model this progress (E4). The central estimates fall inside the 2x–8x band the criteria set, so the hedged '~4x/year' is consistent with credible estimates for around 2022.
Cite this claim
Dario Amodei, “The current rate of algorithmic curve shift is maybe about 4x per year.” (said 13 October 2022). Verdict: resolved true. Reliability API, claim c_4c608c4cad0e, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_4c608c4cad0e
Claims without a verdict
- Unresolvable 157
- Not yet due 40
- Own records only 43
- Not researched 134
- Not checkable 864
Is a verdict here wrong, or do you speak for Dario Amodei? 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.