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.

28,548 claims, sorted by newest said. Choose a claim to read it in full.
VerdictPersonKindSaidClaimTopic
Unresolvable Yann LeCun Factual 8 Mar 2024 Preliminary results suggest V-JEPA representations let the system tell whether a video is physically possible or impossible. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not checkable Yann LeCun Not checkable 8 Mar 2024 Hans Marvek would say that Marvek Paradox still applies. Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 The proportion of all possible prompts used in LLM training is a tiny subset. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not checkable Yann LeCun Not checkable 8 Mar 2024 Well, there’s all kinds of incentive to make AI systems submissive to humans. Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 There is some research on AI planning of tool-use actions, but not a huge amount. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not researched Yann LeCun Factual 8 Mar 2024 Two-level hierarchical planning can be done in AI when the two levels are designed by hand. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Mostly true Yann LeCun Factual 8 Mar 2024 Training a base LLM is currently very expensive and difficult, so only a few companies can do it properly. Yann LeCun · Factual · 8 Mar 2024 Economics and markets
Not checkable Yann LeCun Not checkable 8 Mar 2024 it is very important for people in India. Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 The computation an LLM spends per produced token is constant. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not researched Yann LeCun Factual 8 Mar 2024 Yann LeCun published a paper on contrastive methods in 1993. Yann LeCun · Factual · 8 Mar 2024 Their own work or organisation
Not checkable Yann LeCun Not checkable 8 Mar 2024 Mark is right about a number of things that you list that indeed scare large companies. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 intelligence is not a linear thing that you can measure with a scale or with a single number. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 we don’t want all of our information to come from a unique source because that’s opposite to the whole idea of democracy and progressive… Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 In March 2024, LeCun said he did not think there was yet a good solution for AI planning of tool-use action sequences. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not checkable Yann LeCun Not checkable 8 Mar 2024 in the end, the system ends up being a giant lookup table essentially, which is not really what you want, you want systems that can reason… Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 LLMs behave properly on trained prompts, but a prompt outside the training set can break them and produce nonsense. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Unresolvable Yann LeCun Prediction 8 Mar 2024 In the future, every interaction people have with the digital world will be mediated by their AI assistant. Yann LeCun · Prediction · 8 Mar 2024 AI safety and alignment
Not checkable Yann LeCun Not checkable 8 Mar 2024 But the point is, if you have a big enough potential customer base and you need to build that system anyway for them, it doesn’t hurt you… Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 V-JEPA is Meta's first system to learn video representations good enough for a supervised classifier head to identify actions in video with pretty good accuracy. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Not checkable Yann LeCun Not checkable 8 Mar 2024 And in fact, a lot of doomers are doomers because they don’t think that people are fundamentally good, and they either don’t trust people… Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 It’s going to be you push it in one way, one set of people are going to see it as biased, and then you push it the other way and another… Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 There has to be some limit to the kind of stuff that those systems might be authorized to produce, some guardrails. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 And I call this a JEPA, so that means joint embedding predictive architecture Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 The Ottoman Empire banned the printing press in part to protect the powerful guild of scribes. Yann LeCun · Factual · 8 Mar 2024 Technology
Not checkable Yann LeCun Not checkable 8 Mar 2024 So I’m super excited because I see a path towards potentially human-level intelligence with systems that can understand the world… Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 The printing press enabled philosophy, rationalism, escape from religious doctrine, democracy and science. Yann LeCun · Factual · 8 Mar 2024 Philosophy, ethics and society
Not checkable Yann LeCun Not checkable 8 Mar 2024 It has to do with the acceptability or political leanings of their customer base or audience. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 those are four essential characteristics of intelligent systems or entities, humans, animals. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 So intelligence is a collection of skills and an ability to acquire new skills efficiently. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 No, of course everybody would be impressed. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 And they do this optimization in a horribly inefficient way, which is generate a lot of hypothesis and then select the best ones. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 Well, eventually, yes. But I think if we do this too early, we run the risk of being tempted to cheat. Yann LeCun · Not checkable · 8 Mar 2024
Unresolvable Yann LeCun Prediction 8 Mar 2024 Humanity will not be exposed to a single rogue AI that kills everyone. Yann LeCun · Prediction · 8 Mar 2024 AI safety and alignment
Not checkable Yann LeCun Not checkable 8 Mar 2024 The idea somehow that we can’t get it slightly wrong because if we get it slightly wrong, we’ll die is ridiculous. Yann LeCun · Not checkable · 8 Mar 2024
Not checkable Yann LeCun Not checkable 8 Mar 2024 So there’s going to be various versions of LLaMA that are improvements of previous LLaMAs, bigger, better, multimodal, things like that. Yann LeCun · Not checkable · 8 Mar 2024
Not researched Yann LeCun Factual 8 Mar 2024 Everything humans learn in the first few years of life, and everything animals learn, has nothing to do with language. Yann LeCun · Factual · 8 Mar 2024 AI capabilities
Unresolvable Yann LeCun Prediction 8 Mar 2024 Progress toward human-level AI will be gradual. Yann LeCun · Prediction · 8 Mar 2024 AI timelines and forecasting
Unresolvable Eliezer Yudkowsky Prediction 7 Mar 2024 A trend that looks like an intelligence test will soon start excluding some actual humans, if it hasn't already, and that will lead to an unpleasant conversation. Eliezer Yudkowsky · Prediction · 7 Mar 2024 AI capabilities
Not checkable Eliezer Yudkowsky Not checkable 7 Mar 2024 Less pretty than conversations about difficulty reading weird letters; this *looks* like an intelligence test. Eliezer Yudkowsky · Not checkable · 7 Mar 2024
Not checkable Eliezer Yudkowsky Not checkable 7 Mar 2024 Even this incredibly disastrous chemistry company was vastly more on the ball than current AI developers are on about aligning… Eliezer Yudkowsky · Not checkable · 7 Mar 2024
Mixed Eliezer Yudkowsky Prediction 5 Mar 2024 If a rival AI developer beats GPT-4 on general intelligence, OpenAI will be forced to release GPT-4.5. Eliezer Yudkowsky · Prediction · 5 Mar 2024 AI capabilities
Not checkable Eliezer Yudkowsky Not checkable 5 Mar 2024 It's not cool to compete on general intelligence level. Eliezer Yudkowsky · Not checkable · 5 Mar 2024
Not checkable Eliezer Yudkowsky Not checkable 5 Mar 2024 It's on the capability side of the alignment/capability divide. Eliezer Yudkowsky · Not checkable · 5 Mar 2024
Not checkable Eliezer Yudkowsky Not checkable 5 Mar 2024 Suppose an AI passes the unrestricted text-based Turing Test for IQ 100, but not IQ 115. Eliezer Yudkowsky · Not checkable · 5 Mar 2024
Unresolvable Eliezer Yudkowsky Prediction 5 Mar 2024 AI will appear more helpful, being better at predicting what humans want to see, as AI becomes more capable. Eliezer Yudkowsky · Prediction · 5 Mar 2024 AI safety and alignment
Not checkable Eliezer Yudkowsky Not checkable 3 Mar 2024 To understand the user interfaces on microwave ovens, you need to understand that microwave UI designers are aliens. Eliezer Yudkowsky · Not checkable · 3 Mar 2024
Not checkable Eliezer Yudkowsky Not checkable 3 Mar 2024 There's only so much real estate in a microwave control panel; it's much more important to have an inscrutable button that says "Potato"… Eliezer Yudkowsky · Not checkable · 3 Mar 2024
Not checkable Eliezer Yudkowsky Not checkable 3 Mar 2024 Nobody knows why they insist on hollowing out and inhabiting human appliance designers in particular. Eliezer Yudkowsky · Not checkable · 3 Mar 2024
Mostly false Jensen Huang Factual Mar 2024 Nvidia's Earth digital twin AI model, once trained, can predict weather 3,000 times faster than a supercomputer using 1,000 times less energy. Jensen Huang · Factual · Mar 2024 Science and research methods
Resolved true Jensen Huang Factual Mar 2024 As of March 2024, Google and the whole industry were preparing to adopt Nvidia's Blackwell chip. Jensen Huang · Factual · Mar 2024 AI compute, hardware and data
Not checkable Demis Hassabis Not checkable 28 Feb 2024 The amount of information and models that can be built up just from language is surprising. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 Yes, exactly. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think we get some grounding through the RLHF feedback systems because obviously the human raters are by definition, grounded people. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 In my opinion, they’ve clearly got some form of concepts and abstractions and things like that. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I’m obviously a huge techno-optimist but I want us to be cautious given the transformative power of what we’re bringing into the world… Demis Hassabis · Not checkable · 28 Feb 2024
Unresolvable Demis Hassabis Prediction 28 Feb 2024 Demis Hassabis said he thinks the quickest and most plausible route to AGI is to use existing world knowledge collected from sources like the Web. Demis Hassabis · Prediction · 28 Feb 2024 AI timelines and forecasting
Unresolvable Demis Hassabis Factual 28 Feb 2024 Before founding DeepMind, its founders already knew that progress on these ideas would create unbelievably transformative technology. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Not researched Demis Hassabis Factual 28 Feb 2024 DeepMind's pioneering work, inspired by neuroscience, included experience replay and the notion of attention. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think if it’s not properly grounded, the system won’t be able to achieve those goals properly. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 They’re not at the level where they can create the hypothesis themselves or ask the right question. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 The first part is making sure you detect it ahead of time. Demis Hassabis · Not checkable · 28 Feb 2024
Not researched Demis Hassabis Factual 28 Feb 2024 In 2010, no one was thinking about AI, let alone AGI. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Not checkable Demis Hassabis Not checkable 28 Feb 2024 So I think we are going to need some automated testing. Demis Hassabis · Not checkable · 28 Feb 2024
Own records only Demis Hassabis Factual 28 Feb 2024 Google DeepMind devotes about half its effort to scaling and half to inventing next architectures and algorithms. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Not checkable Demis Hassabis Not checkable 28 Feb 2024 That’s what I think true multimodality is. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I’m very optimistic about that. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think we’re still in the early days of those kinds of systems being at risk. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 You don’t want to be live A/B testing out in the world with these very consequential systems because unintended consequences may be quite… Demis Hassabis · Not checkable · 28 Feb 2024
Own records only Demis Hassabis Factual 28 Feb 2024 Hassabis said that "we" have a lot of processes in place to make sure that only people who need access have access. Demis Hassabis · Factual · 28 Feb 2024 AI safety and alignment
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think that maybe in the next three, four, five years, we would also want air gaps and various other things that are known in the security… Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think no one knows. Demis Hassabis · Not checkable · 28 Feb 2024
Mostly true Demis Hassabis Factual 28 Feb 2024 DeepMind has always bet on generality and learning as the core of its techniques, focusing on reinforcement learning, search and deep learning. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Not checkable Demis Hassabis Not checkable 28 Feb 2024 The brain is an existence proof that general intelligence is possible at all. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 One order of magnitude is probably about the maximum that you want to do between each era. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 In a way, I look at the large models today and I think they’re almost unreasonably effective for what they are. Demis Hassabis · Not checkable · 28 Feb 2024
Mostly true Demis Hassabis Factual 28 Feb 2024 Demis Hassabis thinks human brains plan moves differently from AlphaGo because brains are not built for Monte Carlo tree search. Demis Hassabis · Factual · 28 Feb 2024 Science and research methods
Unresolvable Demis Hassabis Prediction 28 Feb 2024 Self-play ideas could be used to automatically test boundary conditions of new AI systems. Demis Hassabis · Prediction · 28 Feb 2024 AI safety and alignment
Not checkable Demis Hassabis Not checkable 28 Feb 2024 It’s been fantastic actually, over the last year. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think we’ve got to push scaling as hard as we can and that’s what we’re doing here. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 It is more challenging because of the data problem. Demis Hassabis · Not checkable · 28 Feb 2024
Unresolvable Demis Hassabis Prediction 28 Feb 2024 Large predictive world models are a necessary but probably insufficient component of an AGI system. Demis Hassabis · Prediction · 28 Feb 2024 AI timelines and forecasting
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I haven’t heard a compelling, clear answer to that from proponents of just open sourcing everything. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 So that’s very useful already for many of the problems we deal with today, but not the most high-level creative problems. Demis Hassabis · Not checkable · 28 Feb 2024
Research failed Demis Hassabis Factual 28 Feb 2024 Early signs show large models becoming transferable to robotics. Demis Hassabis · Factual · 28 Feb 2024 AI capabilities
Not researched Demis Hassabis Factual 28 Feb 2024 Since DeepMind's founding, it has said that it does not have to wait for AGI to bring incredible benefits to the world. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Mostly true Demis Hassabis Factual 28 Feb 2024 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. Demis Hassabis · Factual · 28 Feb 2024 AI capabilities
Unresolvable Demis Hassabis Prediction 28 Feb 2024 Demis Hassabis's bet is that the final AGI system will include large multimodal models, but that they probably won't be enough alone and will need added planning and search on top. Demis Hassabis · Prediction · 28 Feb 2024 AI timelines and forecasting
Not researched Demis Hassabis Factual 28 Feb 2024 DeepMind joined forces with Google in 2014. Demis Hassabis · Factual · 28 Feb 2024 Their own work or organisation
Unresolvable Demis Hassabis Factual 28 Feb 2024 Google DeepMind has covered the secure-model requirement through world-class cybersecurity and monitoring. Demis Hassabis · Factual · 28 Feb 2024 AI safety and alignment
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think they’re not reliable yet enough to do things like science with them. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 So I think the world’s about to become very exciting in the next few years as we start getting used to the idea of what true multimodality… Demis Hassabis · Not checkable · 28 Feb 2024
Not researched Demis Hassabis Factual 28 Feb 2024 When large AI models improve at coding, their general reasoning can also improve. Demis Hassabis · Factual · 28 Feb 2024 AI capabilities
Unresolvable Demis Hassabis Prediction 28 Feb 2024 In Hassabis's opinion, the AI systems around at the time were not dangerous, but in a few years they might have the potential to be. Demis Hassabis · Prediction · 28 Feb 2024 AI safety and alignment
Not researched Demis Hassabis Factual 28 Feb 2024 RL agents like AlphaZero and AlphaGo are active learners whose decisions determine what data or experience they get next. Demis Hassabis · Factual · 28 Feb 2024 AI capabilities
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think that suggests there must be high-level common algorithmic themes around how the brain processes the world around us. Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think we’re just at the beginning of actually understanding how exciting that might be to interact with a full multimodal model system. Demis Hassabis · Not checkable · 28 Feb 2024
Not researched Demis Hassabis Factual 28 Feb 2024 Downstream capabilities such as MMLU or math performance sometimes don't follow from predictable improvements in training loss. Demis Hassabis · Factual · 28 Feb 2024 Science and research methods
Not checkable Demis Hassabis Not checkable 28 Feb 2024 That pushes us in very interesting research directions that we think are going to be useful anyway: sampling efficiency and data efficiency… Demis Hassabis · Not checkable · 28 Feb 2024
Not checkable Demis Hassabis Not checkable 28 Feb 2024 I think this has to be a big collaboration with many stakeholders from civil society, academia, government, etc. Demis Hassabis · Not checkable · 28 Feb 2024
Not yet due Demis Hassabis Prediction 28 Feb 2024 Hassabis said that DeepMind, started in 2010, was conceived as a 20-year project, which points to about 2030, and that he thinks it is on track. Demis Hassabis · Prediction · 28 Feb 2024 AI timelines and forecasting
Back to the list

Not checkable

“I’m very optimistic about that.”

Demis Hassabis 28 February 2024 Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold (Dwarkesh Podcast)

I’m very optimistic about that.

Not checkable An opinion, value judgement or statement with nothing to check.

Cite this claim

Demis Hassabis, “I’m very optimistic about that.” (said 28 February 2024). Verdict: not checkable. Reliability API, claim c_1b9d809f4c6a, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_1b9d809f4c6a

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