| Verdict | Person | Kind | Said | Claim | Topic |
|---|---|---|---|---|---|
| Resolved true | Ray Dalio | Factual | 9 May 2026 | About one-third of US debt is held by foreigners. Ray Dalio · Factual · 9 May 2026 | Economics and markets |
| Mostly true | Jensen Huang | Factual | 17 Mar 2026 | Satya Nadella confirmed that the first Vera Rubin rack is running on Microsoft Azure. Jensen Huang · Factual · 17 Mar 2026 | Economics and markets |
| Mixed | Jensen Huang | Factual | Nov 2025 | Cloud GPUs are sold out, as of NVIDIA's third quarter fiscal 2026 results (Nov. 2025). Jensen Huang · Factual · Nov 2025 | AI compute, hardware and data |
| Mixed | Sam Altman | Factual | 9 Aug 2025 | GPT-5 is significantly better than earlier models at healthcare-related queries. Sam Altman · Factual · 9 Aug 2025 | AI capabilities |
| Resolved true | Gary Marcus | Prediction | 27 Jul 2025 | If the model being discussed has guardrails like ChatGPT's, they will teeter between being too weak (beaten by jailbreaks) and too strong (rejecting some perfectly reasonable requests). Gary Marcus · Prediction · 27 Jul 2025 | AI safety and alignment |
| Mostly true | Dario Amodei | Factual | 18 Feb 2025 | The lack of understanding of how AI systems work is essentially unprecedented in the history of technology. Dario Amodei · Factual · 18 Feb 2025 | AI safety and alignment |
| Resolved true | Rodney Brooks | Factual | 1 Jan 2025 | Despite further thruster anomalies, Starliner would have landed safely had the crew been aboard when it returned empty in 2024. Rodney Brooks · Factual · 1 Jan 2025 | Economics and markets |
| Resolved true | Tyler Cowen | Factual | 25 Dec 2024 | The war in Ethiopia affecting Yonas's region is over in the immediate sense. Tyler Cowen · Factual · 25 Dec 2024 | Other |
| Mostly true | Peter Thiel | Factual | Apr 2024 | René Girard often said that Christ was the first political atheist. Peter Thiel · Factual · Apr 2024 | Philosophy, ethics and society |
| Mostly true | Elon Musk | Factual | 19 Mar 2024 | Political donations from old Twitter were 99% to Democrats. Elon Musk · Factual · 19 Mar 2024 | Philosophy, ethics and society |
| Mostly true | Marc Andreessen | Factual | 12 Jun 2023 | As of June 2023, the U.S. government was both alarmed about deepfakes and trying to outlaw blockchains. Marc Andreessen · Factual · 12 Jun 2023 | AI policy and governance |
| Mostly true | Elon Musk | Promise | 26 Jan 2023 | Tesla expects to use the Dojo computer operationally later in 2023. Elon Musk · Promise · 26 Jan 2023 | AI compute, hardware and data |
| Mostly true | Jeremy Grantham | Factual | 31 Aug 2022 | Few superbubbles exist in history to study, and they share clear common features. Jeremy Grantham · Factual · 31 Aug 2022 | Economics and markets |
| Resolved true | Daniela Amodei | Factual | 2022 | As of 2022 there are no clear guidelines for AI developers on how to avoid advancing race dynamics or do the right thing on safety. Daniela Amodei · Factual · 2022 | AI policy and governance |
| Resolved true | Daniela Amodei | Factual | 2022 | Anthropic trains large-scale generative models and does safety research on those models. Daniela Amodei · Factual · 2022 | Their own work or organisation |
| Mostly true | GiveWell | Promise | 28 May 2020 | In 2020 GiveWell will complete its review of the New Incentives RCT results and decide whether to direct additional funding to New Incentives. GiveWell · Promise · 28 May 2020 | Their own work or organisation |
| Mostly true | Marc Andreessen | Factual | 1 Jan 2018 | Andreessen Horowitz initially made no healthcare-related investments. Marc Andreessen · Factual · 1 Jan 2018 | Their own work or organisation |
| Mostly false | Ray Kurzweil | Prediction | 3 Jun 2015 | Kurzweil predicted that humans will gradually merge with technology and enhance themselves. Ray Kurzweil · Prediction · 3 Jun 2015 | AI timelines and forecasting |
| Mostly true | Nassim Taleb | Factual | 12 Dec 2012 | Theory is born from practice more often than practice from theory, with technology more often preceding science. Nassim Taleb · Factual · 12 Dec 2012 | Science and research methods |
| Resolved true | Jeff Bezos | Factual | 2001 | Amazon.com said it had been told that its American Customer Satisfaction Index score of 84 was the highest ever recorded for a service company in any industry. Jeff Bezos · Factual · 2001 | Economics and markets |
“But I think what that looks like in practice is that we are training large scale generative models, and we're doing safety research on those models.”
Daniela Amodei, Anthropic president 2022 Podcast Transcript: Daniela and Dario Amodei on Anthropic (LessWrong)
Anthropic trains large-scale generative models and does safety research on those models.
Resolved true The record shows it happened or was so.
The second judge said mixed; the verdict shown is the first judge's.
The evidence is the person's or their organisation's own records.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
Judge as of 2022. True if, in 2022, Anthropic was training large-scale generative (language) models and publishing or conducting safety research on those same models, shown for example by 2022 papers on large language models of its own training (such as work on HH-RLHF, red teaming, Constitutional AI, or interpretability). Mostly true if both activities are documented but the safety research used mostly smaller or other models. Mixed if only one of the two activities is documented for 2022. Mostly false if it trained only small models, or only did research on other organisations' models. False if it neither trained large generative models nor did safety research on them in 2022.
Evidence
- Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned arXiv Own record
“First, we investigate scaling behaviors for red teaming across 3 model sizes (2.7B, 13B, and 52B parameters) and 4 model types: a plain language model (LM); an LM prompted to be helpful, honest, and harmless; an LM with rejection sampling; and a model trained to be helpful and harmless using reinforcement learning from human feedback (RLHF).”
- Training a helpful and harmless assistant with RLHF Anthropic · 12 April 2022 Own record
“We apply preference modeling and reinforcement learning from human feedback (RLHF) to finetune language models to act as helpful and harmless assistants.”
- Anthropic's core views on AI safety Anthropic · 8 March 2023 Own record
“We trained the first version of our headline model, Claude, in the spring of 2022, and decided to prioritize using it for safety research rather than public deployments.”
Reasoning
E2, dated April 2022, is Anthropic's published RLHF research, in which it fine-tuned language models to act as helpful and harmless assistants. E1 is a red-teaming paper (arXiv 2209) that studies safety behaviour across language models of up to 52B parameters, including RLHF-trained models. E3 is an official Anthropic statement that it trained the first version of Claude in spring 2022 and chose to use it mainly for safety research. All three sources are Anthropic's own published research or official records, and together they show it trained large generative models in 2022 and did safety research on them.
Cite this claim
Daniela Amodei, “Anthropic trains large-scale generative models and does safety research on those models.” (said 2022). Verdict: resolved true. Reliability API, claim c_dc5120b73b51, method 0.1.0+28da144b. https://epistacklabs.com/claims/c_dc5120b73b51
Verdicts are made automatically; how verdicts and scores are made. The same data is on Hugging Face under CC BY 4.0.