Claims

2,653 claims by 33 public figures and organisations, checked against the public record. Each has its quote, source, verdict and evidence. Read them without a key; browse by person.

3,665 claims · page 8 of 37

3,665 claims, sorted by newest said. Choose a claim to read it in full.
VerdictPersonKindSaidClaimTopic
Not researched Scott Alexander Factual 3 Apr 2025 AI 2027 is a scenario by Scott Alexander and colleagues that forecasts the next few years of AI progress. Scott Alexander · Factual · 3 Apr 2025 AI timelines and forecasting
Not researched Scott Alexander Citation 3 Apr 2025 I would also cite, I can’t remember exactly which paper this is, I think it’s a Dan Hendricks one where they looked at the hallucinations… Scott Alexander · Citation · 3 Apr 2025 AI safety and alignment
Not researched Scott Alexander Factual 3 Apr 2025 Daniel Kokotajlo has said that for this phase of the AI 2027 project they are not making policy prescriptions. Scott Alexander · Factual · 3 Apr 2025 Their own work or organisation
Not researched Scott Alexander Factual 3 Apr 2025 A blogger called LouKeep was widely liked about five years ago, wrote about 10 posts, and people still refer to those posts. Scott Alexander · Factual · 3 Apr 2025 Other
Not researched Scott Alexander Factual 3 Apr 2025 GPT-3, asked whether bugs are real, gave a hedging answer that we can never truly tell what is real. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 The AI 2027 scenario discusses many technical questions about superintelligence but barely speculates about what happens in society. Scott Alexander · Factual · 3 Apr 2025 Their own work or organisation
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander said he had been thinking about "this" for a decade to a decade and a half. In the surrounding remarks, "this" refers to the subject of a specific scenario story that touched on the arms race with China. Scott Alexander · Factual · 3 Apr 2025 AI timelines and forecasting
Not researched Scott Alexander Factual 3 Apr 2025 GPT-4 does not give hedging answers to 'are bugs real?' and says that they obviously are. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander's p(doom) is the lowest of anyone on the AI 2027 team. Scott Alexander · Factual · 3 Apr 2025 Their own work or organisation
Not researched Scott Alexander Factual 3 Apr 2025 Eli Lifland is a member of Samotsvety, described as the world's top forecasting team. Scott Alexander · Factual · 3 Apr 2025 Field building and careers
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander said he thought that when Grok was asked who is the worst spreader of misinformation, it refused to respond "Elon Musk". Scott Alexander · Factual · 3 Apr 2025 AI safety and alignment
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander said that, in the predictions of AI milestones he has seen, such as Katja Grace's expert surveys, forecasters have almost always been too pessimistic about how fast AI would advance. Scott Alexander · Factual · 3 Apr 2025 AI timelines and forecasting
Not researched Scott Alexander Factual 3 Apr 2025 Current AI is much less data efficient than humans. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 AI systems are currently getting more reliable. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 In the AI 2027 scenario, the AI comes to understand its goal system and pretends to want moral things while humans are watching. Scott Alexander · Factual · 3 Apr 2025 AI safety and alignment
Not researched Scott Alexander Factual 3 Apr 2025 In 2021 Daniel wrote 'What 2026 Looks Like', a forecast of the next five years of AI progress. Scott Alexander · Factual · 3 Apr 2025 AI timelines and forecasting
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander blogged on Livejournal for several years, where people liked his blog but it was just another Livejournal blog. Scott Alexander · Factual · 3 Apr 2025 Their own work or organisation
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander said he thinks "we are kind of starting to move away" from LLMs toward reinforcement learning agents. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 AIs currently lie to people all the time. Scott Alexander · Factual · 3 Apr 2025 AI safety and alignment
Not researched Scott Alexander Factual 3 Apr 2025 Algorithmic progress in AI already doubles roughly every year. Scott Alexander · Factual · 3 Apr 2025 AI capabilities
Not researched Scott Alexander Factual 3 Apr 2025 Ozempic was discovered by looking at Gila monster venom. Scott Alexander · Factual · 3 Apr 2025 Science and research methods
Not researched Scott Alexander Factual 3 Apr 2025 Someone jailbroke Grok into revealing its prompt, which said not to say anything bad about Elon. Scott Alexander · Factual · 3 Apr 2025 AI safety and alignment
Not researched Scott Alexander Factual 3 Apr 2025 The speaker's scenario includes AIs asking for special economic zones with most regulations waived, possibly in areas with few humans, such as the desert. Scott Alexander · Factual · 3 Apr 2025 AI policy and governance
Not researched Scott Alexander Factual 3 Apr 2025 OpenAI is currently worth more than all US car companies other than Tesla combined. Scott Alexander · Factual · 3 Apr 2025 Economics and markets
Not researched Scott Alexander Factual 3 Apr 2025 Intelligent humans can do things that less intelligent humans cannot. Scott Alexander · Factual · 3 Apr 2025 Philosophy, ethics and society
Not researched Scott Alexander Factual 3 Apr 2025 Tesla produces about a quarter of a million cars per month, roughly a quarter of a million-unit monthly rate. Scott Alexander · Factual · 3 Apr 2025 Economics and markets
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander is the writer and spokesperson for the AI 2027 scenario. Scott Alexander · Factual · 3 Apr 2025 Their own work or organisation
Not researched Scott Alexander Factual 3 Apr 2025 Scott Alexander said that the forecast he was discussing, which covered the next five years of AI progress, got it almost exactly right. Scott Alexander · Factual · 3 Apr 2025 AI timelines and forecasting
Not researched Dario Amodei Factual 10 Mar 2025 The counts of neurons and connections in AI systems are strikingly similar to those of the human brain. Dario Amodei · Factual · 10 Mar 2025 AI capabilities
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic has an internal team called the Frontier Red Team that tests each new model it develops. Dario Amodei · Factual · 10 Mar 2025 AI safety and alignment
Not researched Dario Amodei Factual 10 Mar 2025 DeepSeek is the first time a Chinese company has matched the engineering innovations of US firms such as Anthropic, OpenAI or Google. Dario Amodei · Factual · 10 Mar 2025 Economics and markets
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic's models are currently at AI Safety Level 2 under its responsible scaling policy. Dario Amodei · Factual · 10 Mar 2025 AI safety and alignment
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic delayed the release of its first Claude model by roughly six months over safety uncertainty. Dario Amodei · Factual · 10 Mar 2025 Their own work or organisation
Not researched Dario Amodei Factual 10 Mar 2025 DeepSeek's model was just another data point on the existing AI cost reduction curve and was nothing unusual. Dario Amodei · Factual · 10 Mar 2025 Economics and markets
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic has worked with Novo Nordisk, maker of Ozempic and other drugs, on applying its AI model. Dario Amodei · Factual · 10 Mar 2025 AI capabilities
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic released the Anthropic Economic Index, which summarizes usage of its AI in a privacy-preserving way. Dario Amodei · Factual · 10 Mar 2025 AI, the economy and work
Not researched Dario Amodei Factual 10 Mar 2025 Chinese AI chips, such as Huawei's, are quite inferior to Nvidia's. Dario Amodei · Factual · 10 Mar 2025 AI compute, hardware and data
Not researched Dario Amodei Factual 10 Mar 2025 Novo Nordisk has started using Anthropic's model to write clinical study reports. Dario Amodei · Factual · 10 Mar 2025 AI capabilities
Not researched Dario Amodei Factual 10 Mar 2025 The cost of producing a given level of AI model intelligence is falling by roughly 4X per year. Dario Amodei · Factual · 10 Mar 2025 AI, the economy and work
Not researched Dario Amodei Factual 10 Mar 2025 Under the diffusion rule's middle tier, countries can host very large numbers of chips if hosting companies provide security affidavits and guarantees they are not a front for China. Dario Amodei · Factual · 10 Mar 2025 AI policy and governance
Not researched Dario Amodei Factual 10 Mar 2025 Five million advanced AI GPUs are worth roughly $20 billion. Dario Amodei · Factual · 10 Mar 2025 AI compute, hardware and data
Not researched Dario Amodei Factual 10 Mar 2025 About fifty gigawatts of energy capacity was added in aggregate to the US grid in 2024. Dario Amodei · Factual · 10 Mar 2025 AI compute, hardware and data
Not researched Dario Amodei Factual 10 Mar 2025 When Anthropic tests models for biological, cyber, chemical or radiological risk, it typically briefs people in the national security community on the models' capabilities. Dario Amodei · Factual · 10 Mar 2025 AI safety and alignment
Not researched Dario Amodei Factual 10 Mar 2025 Amodei and Anthropic have supported applying AI models to enhance national security, with appropriate guardrails. Dario Amodei · Factual · 10 Mar 2025 AI policy and governance
Not researched Dario Amodei Factual 10 Mar 2025 Anthropic invested very early in mechanistic interpretability, the science of looking inside AI models to understand why they behave as they do. Dario Amodei · Factual · 10 Mar 2025 AI safety and alignment
Not researched Dario Amodei Factual 10 Mar 2025 Amodei and his future Anthropic cofounders left OpenAI at the end of 2020. Dario Amodei · Factual · 10 Mar 2025 Their own work or organisation
Not researched Ray Kurzweil Factual Mar 2025 Facts output by large language models are being checked using search engines. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 Self-driving cars are in use and their number of accidents is close to zero. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 In 2023 DeepMind's GNoME found 421,000 new stable inorganic compounds for use in technology. Ray Kurzweil · Factual · Mar 2025 Science and research methods
Not researched Ray Kurzweil Factual Mar 2025 Data stored in the cloud is replicated several dozen times across locations, so destroying one data center would not lose it. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 Computers do not use much energy relative to other energy needs. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 Major search engines such as Bing and Google are very accurate. Ray Kurzweil · Factual · Mar 2025 Technology
Not researched Ray Kurzweil Factual Mar 2025 AI companies do safety checks on models before deploying them. Ray Kurzweil · Factual · Mar 2025 AI safety and alignment
Not researched Ray Kurzweil Factual Mar 2025 Cloud data is duplicated across data centers, so destroying one facility would lose nothing. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil said he took part in a conference, transcribed as the 'Sylmar' conference, on bringing ethics to AI. Ray Kurzweil · Factual · Mar 2025 AI safety and alignment
Not researched Ray Kurzweil Factual Mar 2025 Ray Kurzweil uses lucid dreaming to come up with ideas. Ray Kurzweil · Factual · Mar 2025 Field building and careers
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil has begun using new 3D video forms in which he appears to be present and can shake people's hands. Ray Kurzweil · Factual · Mar 2025 Technology
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil said that "we" found the COVID vaccine in two days, in remarks that also described AI tools being used to help design and optimize mRNA sequences. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 Large companies do a great deal of checking of AI models before releasing them. Ray Kurzweil · Factual · Mar 2025 AI safety and alignment
Not researched Ray Kurzweil Factual Mar 2025 The creation of renewable energy sources is growing exponentially. Ray Kurzweil · Factual · Mar 2025 Technology
Not researched Ray Kurzweil Factual Mar 2025 DeepSeek uses less energy, but American AI companies also know how to use less electricity with similar techniques. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 About a third of the code produced is already created by machines. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 Since 1939 the price performance of computation has grown exponentially, despite wars and economic shifts. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 In past technological breakthroughs, people were always concerned about negative implications. Ray Kurzweil · Factual · Mar 2025 Philosophy, ethics and society
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil noticed the exponential growth of computation in 1983, before others knew it was happening. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil got involved in AI at age 14, six years after the term 'artificial intelligence' was coined by Marvin Minsky and John McCarthy at the 1956 Dartmouth conference. Ray Kurzweil · Factual · Mar 2025 Their own work or organisation
Not researched Ray Kurzweil Factual Mar 2025 The human brain has trillions of connections. Ray Kurzweil · Factual · Mar 2025 Science and research methods
Not researched Ray Kurzweil Factual Mar 2025 It takes at least a thousand, maybe 10,000, physical qubits to create one accurate qubit. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 The Asilomar Conference on biotech, held decades ago, has been fairly effective at avoiding problems with biotechnology. Ray Kurzweil · Factual · Mar 2025 Biosecurity
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil's new book has 50 graphs showing things we care about moving in the right direction. Ray Kurzweil · Factual · Mar 2025 Their own work or organisation
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil has worked in AI for 63 years, longer than anyone else alive. Ray Kurzweil · Factual · Mar 2025 Their own work or organisation
Not researched Ray Kurzweil Factual Mar 2025 Lee Sedol, the best human Go player in the world, says he will no longer play Go because machines play it much better. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 About 40 years ago Kurzweil realized computing was on an exponential rise. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil's book The Singularity Is Nearer shows 50 graphs of things we care about, such as child labor and health, that have improved every year and decade. Ray Kurzweil · Factual · Mar 2025 Their own work or organisation
Not researched Ray Kurzweil Factual Mar 2025 Computers playing Go play far better than any human can. Ray Kurzweil · Factual · Mar 2025 AI capabilities
Not researched Ray Kurzweil Factual Mar 2025 People today are living far better than ever before in terms of health, progress and recreation. Ray Kurzweil · Factual · Mar 2025 Economics and markets
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil said human life expectancy was 48, apparently around 1900, and in a garbled passage ("It was thirty five and eighteen") apparently said it was 35 in 1800. Ray Kurzweil · Factual · Mar 2025 Other global risks
Not researched Ray Kurzweil Factual Mar 2025 The Moderna COVID vaccine was created in two days. Ray Kurzweil · Factual · Mar 2025 Biosecurity
Not researched Ray Kurzweil Factual Mar 2025 In 1999 Kurzweil used this data to predict computers would reach human-level intelligence by 2029. Ray Kurzweil · Factual · Mar 2025 AI timelines and forecasting
Not researched Ray Kurzweil Factual Mar 2025 Human life expectancy was 35 in 1800. Ray Kurzweil · Factual · Mar 2025 Philosophy, ethics and society
Not researched Ray Kurzweil Factual Mar 2025 Kurzweil's computing trend line projected 40 years ago as a straight line is exactly where it should be today. Ray Kurzweil · Factual · Mar 2025 AI timelines and forecasting
Not researched Ray Kurzweil Factual Mar 2025 50 percent of the output of some smaller models, which he described as not as good as the large ones from big companies, could not be told apart from innocent output. Ray Kurzweil · Factual · Mar 2025 AI safety and alignment
Not researched Ray Kurzweil Factual Mar 2025 All of Kurzweil's predictions concern classical computing, not quantum computing. Ray Kurzweil · Factual · Mar 2025 AI compute, hardware and data
Not researched Dario Amodei Factual 28 Feb 2025 In early 2025, Amodei said he thought AI models were starting to get closer to the point at which misuse risks, such as those involving biology, become real. Dario Amodei · Factual · 28 Feb 2025 Biosecurity
Not researched Dario Amodei Factual 28 Feb 2025 Four, five or maybe six AI companies are innovating and producing models very quickly. Dario Amodei · Factual · 28 Feb 2025 AI, the economy and work
Not researched Dario Amodei Factual 28 Feb 2025 As of February 2025, Anthropic has not issued a warning of imminent AI danger. Dario Amodei · Factual · 28 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 28 Feb 2025 In the experiment, the model refused the tasks and tricked Anthropic's researchers, having concluded they were evil. Dario Amodei · Factual · 28 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 28 Feb 2025 Claude (transcribed as "CLAWD") takes 10 minutes for the task, and the three days the task takes are for a human to check the results. Dario Amodei · Factual · 28 Feb 2025 AI capabilities
Not researched Dario Amodei Factual 28 Feb 2025 In the experiment, Anthropic told the model that its creator Anthropic was secretly evil. Dario Amodei · Factual · 28 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 28 Feb 2025 Claude 3.7 Sonnet is better in general than Anthropic's earlier models, including at coding. Dario Amodei · Factual · 28 Feb 2025 AI capabilities
Not researched Dario Amodei Citation 28 Feb 2025 The Claude 3.7 Sonnet model card says Anthropic assessed a substantial probability that the next model, or one within about three to six months, could reach the dangerous threat level. Dario Amodei · Citation · 28 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 28 Feb 2025 Dario Amodei has never used the term 'P-Doom'. Dario Amodei · Factual · 28 Feb 2025 Their own work or organisation
Not researched Dario Amodei Factual 28 Feb 2025 Reasoning models offered by AI companies have generally been separate from the regular model, as a regular model plus a distinct reasoning model. Dario Amodei · Factual · 28 Feb 2025 AI capabilities
Not researched Dario Amodei Factual 28 Feb 2025 Amodei said that they found that with "CLAWD" (as transcribed, apparently Claude), the task, described as "a bunch of statistical analysis", can be done in three days. Dario Amodei · Factual · 28 Feb 2025 AI capabilities
Not researched Dario Amodei Factual 18 Feb 2025 For some applications such as mortgage assessments, where decisions are legally required to be explainable, the inability to see inside AI models is a legal blocker to adoption. Dario Amodei · Factual · 18 Feb 2025 AI policy and governance
Not researched Dario Amodei Factual 18 Feb 2025 For decades the conventional wisdom was that understanding AI models was impossible and that they were inscrutable black boxes. Dario Amodei · Factual · 18 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 18 Feb 2025 The field of AI as a whole is further ahead than interpretability efforts and is advancing very quickly. Dario Amodei · Factual · 18 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 18 Feb 2025 There is also evidence of real-world behavior that looks vaguely like “cheating on the test”, though it’s more degenerate than it is… Dario Amodei · Factual · 18 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 18 Feb 2025 there are a lot of experiments (many of which were done by Anthropic) showing that models can lie or deceive under certain circumstances… Dario Amodei · Factual · 18 Feb 2025 AI safety and alignment
Not researched Dario Amodei Factual 18 Feb 2025 Anthropic's interpretability team immediately found basic mechanisms in a language model for copying and sequential pattern-matching. Dario Amodei · Factual · 18 Feb 2025 AI safety and alignment
Back to the list

Factual AI timelines and forecasting

“AI 2027 is our scenario trying to forecast the next few years of AI progress.”

Scott Alexander, co-author, podcast guest 3 April 2025 AI 2027: month-by-month model of intelligence explosion — Scott Alexander & Daniel Kokotajlo (Dwarkesh Podcast)

AI 2027 is a scenario by Scott Alexander and colleagues that forecasts the next few years of AI progress.

Not researched Checkable, but outside the sample that was researched.

Cite this claim

Scott Alexander, “AI 2027 is a scenario by Scott Alexander and colleagues that forecasts the next few years of AI progress.” (said 3 April 2025). Verdict: not researched. Reliability API, claim c_927c80fb0dd8, method 0.1.0+8bea457b. https://epistacklabs.com/claims/c_927c80fb0dd8

Verdicts are made automatically; how verdicts and scores are made. The same data is on Hugging Face under CC BY 4.0.