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
Resolved true Jeff Bezos Factual Apr 2019 Third-party sellers' share of physical gross merchandise sales on Amazon has grown from 3% of the total to 58%. Jeff Bezos · Factual · Apr 2019 Economics and markets
Not checkable Jeff Bezos Not checkable Apr 2019 We must invent on their behalf. Jeff Bezos · Not checkable · Apr 2019
Resolved true Jeff Bezos Factual Apr 2019 Amazon began developing the Fire phone and the Echo around the same time. Jeff Bezos · Factual · Apr 2019 Technology
Not checkable Jeff Bezos Not checkable Apr 2019 We’re also plunging into helping companies harness Machine Learning. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 As a company grows, everything needs to scale, including the size of your failed experiments. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Teams all across Amazon are listening to customers and wandering on their behalf! Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Much of what we build at AWS is based on listening to customers. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 From very early on in Amazon’s life, we knew we wanted to create a culture of builders – people who are curious, Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Getting there was hard. Technically hard. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 As always, I attach a copy of our original 1997 letter. It remains Day 1. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Sometimes (often actually) in business, you do know where you’re going, and when you do, you can be efficient. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Third-party sellers are kicking our first party butt. Badly. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Many of the other programs we have introduced for our employees came as much from the heart as the head. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Within AWS, that same pattern has recurred many times. Jeff Bezos · Not checkable · Apr 2019
Unresolvable Jeff Bezos Factual Apr 2019 Amazon invested in and offered independent sellers selling tools such as inventory management, payments, shipment tracking, reporting and cross-border selling. Jeff Bezos · Factual · Apr 2019 Economics and markets
Not checkable Jeff Bezos Not checkable Apr 2019 But we’re also optimistic about specialized databases for specialized workloads. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 No one likes to wait in line. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 No customer was asking for Echo. This was definitely us wandering. Market research doesn’t help. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 But that is not what drove the decision. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Something strange and remarkable has happened over the last 20 years. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 SageMaker removes the heavy lifting, complexity, and guesswork from each step of the machine learning Jeff Bezos · Not checkable · Apr 2019
Mostly true Jeff Bezos Factual Apr 2019 Third-party sales on Amazon grew from $0.1 billion to $160 billion over the period discussed, a compound annual growth rate of 52%. Jeff Bezos · Factual · Apr 2019 Economics and markets
Mostly true Jeff Bezos Factual Apr 2019 Since the first-generation Echo, customers have purchased more than 100 million Alexa-enabled devices. Jeff Bezos · Factual · Apr 2019 Technology
Not checkable Jeff Bezos Not checkable Apr 2019 AWS itself – as a whole – is an example. No one asked for AWS. No one. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 The challenges and hurdles we will face to make our long-term vision for Amazon.com a reality are several: aggressive, capable, well-funded… Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 possible without a culture of curiosity and a willingness to try totally new things on behalf of customers. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Of course, we won’t undertake such experiments cavalierly. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Though we are optimistic, we must remain vigilant and maintain a sense of urgency. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 We are still in the early stages of learning how to bring new value to our customers through Internet Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 counter-balance to efficiency. You need to employ both. Jeff Bezos · Not checkable · Apr 2019
Not checkable Jeff Bezos Not checkable Apr 2019 Better yet, go to $16 and throw the gauntlet back at us. Jeff Bezos · Not checkable · Apr 2019
Mostly true Jeff Bezos Factual Apr 2019 Amazon represents a low single-digit percentage of the global retail market, with much larger retailers in every country where it operates. Jeff Bezos · Factual · Apr 2019 Economics and markets
Not checkable Ray Dalio Not checkable 31 Mar 2019 If you’re looking for a visionary, pick a visionary to do the interview in which you probe for vision. Ray Dalio · Not checkable · 31 Mar 2019
Not checkable Gary Marcus Not checkable 31 Mar 2019 Scary and important work on AI and cybersecurity: Gary Marcus · Not checkable · 31 Mar 2019
Not checkable Ray Dalio Not checkable 31 Mar 2019 If you are looking for a mix of qualities, assemble a group of interviewers who embody those qualities collectively. Ray Dalio · Not checkable · 31 Mar 2019
Not checkable Gary Marcus Not checkable 31 Mar 2019 shows Tesla system lacks adequate redundancy and common sense. Gary Marcus · Not checkable · 31 Mar 2019
Not checkable Ray Dalio Not checkable 30 Mar 2019 Personality assessments are valuable tools for getting a quick picture of what people are like in terms of their abilities, preferences… Ray Dalio · Not checkable · 30 Mar 2019
Not checkable Ray Dalio Not checkable 30 Mar 2019 They are often more objective and reliable than interviews. Ray Dalio · Not checkable · 30 Mar 2019
Not checkable Ray Dalio Not checkable 29 Mar 2019 Some qualities are more suitable for some jobs. Ray Dalio · Not checkable · 29 Mar 2019
Not checkable Gary Marcus Not checkable 28 Mar 2019 For anyone who thought I was jousting at strawmen Gary Marcus · Not checkable · 28 Mar 2019
Not checkable Ray Dalio Not checkable 28 Mar 2019 It is unacceptable to use your personal influence to help someone get a job because doing so undermines the meritocracy. Ray Dalio · Not checkable · 28 Mar 2019
Not checkable Ray Dalio Not checkable 27 Mar 2019 Too many people get hired because they are just “one of those.” Ray Dalio · Not checkable · 27 Mar 2019
Not checkable Ray Dalio Not checkable 25 Mar 2019 The process for choosing people should be systematically built out and evidence-based. Ray Dalio · Not checkable · 25 Mar 2019
Not checkable Ray Dalio Not checkable 25 Mar 2019 Hire right because the penalties for hiring wrong are huge Ray Dalio · Not checkable · 25 Mar 2019
Not checkable Ray Dalio Not checkable 23 Mar 2019 When building a “machine,” design precedes people because the type of people you will need will depend on the design. Ray Dalio · Not checkable · 23 Mar 2019
Not checkable Ray Dalio Not checkable 22 Mar 2019 The most important thing you can do is understand that hiring is a high-risk gamble that needs to be approached deliberately. Ray Dalio · Not checkable · 22 Mar 2019
Not checkable Ray Dalio Not checkable 22 Mar 2019 Recognize that the most important decision for you to make is who you choose as your responsible parties Ray Dalio · Not checkable · 22 Mar 2019
Not checkable Ray Dalio Not checkable 21 Mar 2019 In most cases, those forces were specific people with specific qualities who worked in specific ways. Ray Dalio · Not checkable · 21 Mar 2019
Not checkable Ray Dalio Not checkable 20 Mar 2019 If the owners are self-funded, they still have to make their clients and employees happy. Ray Dalio · Not checkable · 20 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 Saying that in the future we will have viable training sets is shifting the human workload to creating massive training sets and encoding… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 the lesson we should learn from the last seventy years of AI research is not at all that we should just use more computation and that… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 It is sleight of hand in moving the human intellectual work to somewhere else. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 it once again is simply shifting where human intelligence needs to be applied to make ML practical, not eliminating the need for humans to… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 That is not a sustainable model for getting further in intelligent systems. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 Once again it is sleight of hand to say that AI succeeds without humans getting into the loop. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 To have a Deep Learning network also have to learn that seems pedantic to the extreme, and will drive up the computational costs of the… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 a better lesson to be learned is that we have to take into account the total cost of any solution, and that so far they have all required… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 So Sutton’s argument just makes this worse, and makes the use of AI and ML impractical. Rodney Brooks · Not checkable · 19 Mar 2019
Not researched Rodney Brooks Factual 19 Mar 2019 The front end of a CNN is designed by humans to handle translational invariance, so objects can appear anywhere in the frame. Rodney Brooks · Factual · 19 Mar 2019 AI capabilities
Resolved true Rodney Brooks Factual 19 Mar 2019 Computer architects are building special-purpose chips for runtime use of trained neural networks to compensate for power and performance limits. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not researched Rodney Brooks Factual 19 Mar 2019 For most machine learning problems today, a human must design a specific network architecture for learning to proceed well. Rodney Brooks · Factual · 19 Mar 2019 AI capabilities
Not researched Rodney Brooks Factual 19 Mar 2019 Self-driving cars require about 2,500 Watts of power for computation. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Rodney Brooks Not checkable 19 Mar 2019 No human makes that error because they know that stop signs are red, and speed limit signs are white. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 First it locks down hardware specific to particular solutions, so every time we have a new ML problem we will need to design new hardware. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 There are other things in image labeling that suffer mightily because the current crop of CNNs do not have certain things built in that we… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 I think Sutton is wrong for a number of reasons. Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Jeff Bezos Not checkable 19 Mar 2019 Thank you @FestoAG, but bigger please…. I want to get on! #ThinkBig #MARS2019 Jeff Bezos · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 This resonates with a current mode of thinking among many of the newer entrants to AI that it is better to design learning networks and put… Rodney Brooks · Not checkable · 19 Mar 2019
Unresolvable Rodney Brooks Factual 19 Mar 2019 As of 2019, the computation needed to train many current neural networks can only be provided by very large companies with very large budgets. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not researched Rodney Brooks Factual 19 Mar 2019 A human brain requires only about 20 Watts of power. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Rodney Brooks Not checkable 19 Mar 2019 Rather we are asking the humans to pour their intelligence into the algorithms in a different place and form. Rodney Brooks · Not checkable · 19 Mar 2019
Not researched Rodney Brooks Factual 19 Mar 2019 For some machine learning problems, a significant carbon footprint is starting to be seen because of the power consumed in the learning phase. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Ray Dalio Not checkable 19 Mar 2019 So long as you bear the consequences of failure, you are the ultimate Responsible Party. Ray Dalio · Not checkable · 19 Mar 2019
Mostly true Rodney Brooks Factual 19 Mar 2019 The push to make everything learnable is raising the cost of AI beyond what individuals or even large university departments can afford. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Rodney Brooks Not checkable 19 Mar 2019 Saying that a particular solution style minimizes a particular sort of human ingenuity that is needed while not taking into account all the… Rodney Brooks · Not checkable · 19 Mar 2019
Not checkable Rodney Brooks Not checkable 19 Mar 2019 This is just as much building knowledge in as it would be to directly build a color constancy stage. Rodney Brooks · Not checkable · 19 Mar 2019
Unresolvable Rodney Brooks Factual 19 Mar 2019 Brooks says his review, including this comment, is seventy-six words shorter than Sutton's post. Rodney Brooks · Factual · 19 Mar 2019 Their own work or organisation
Unresolvable Rodney Brooks Factual 19 Mar 2019 At a two-day Deep Learning workshop at the National Academy of Science in the week before 2019-03-19, the view favouring task-specialised computation structures was more in vogue than pure learning with massive compute. Rodney Brooks · Factual · 19 Mar 2019 AI capabilities
Not checkable Rodney Brooks Not checkable 19 Mar 2019 The CNN doesn’t know that, because the relationship between pixel color in the camera and the actual color of the object is a very complex… Rodney Brooks · Not checkable · 19 Mar 2019
Mostly true Rodney Brooks Factual 19 Mar 2019 Moore's Law is slowing, and some computer architects report the doubling time of computation on a single chip moving from one year to twenty years. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Rodney Brooks Not checkable 19 Mar 2019 Massive data sets are not at all what humans need to learn things so something is missing. Rodney Brooks · Not checkable · 19 Mar 2019
Not researched Rodney Brooks Factual 19 Mar 2019 Dennard scaling broke down in 2006, so machine power consumption rises as performance improves, making it unaffordable to run machine learning results on many small robots. Rodney Brooks · Factual · 19 Mar 2019 AI compute, hardware and data
Not checkable Ray Dalio Not checkable 19 Mar 2019 If you assign the wrong person to achieve the goal and suffer the consequences, it’s your fault, not the person who failed’s fault. Ray Dalio · Not checkable · 19 Mar 2019
Resolved true Rodney Brooks Factual 19 Mar 2019 Today's machine learning data sets can contain billions of examples, whereas a human may need only a handful to learn the same thing. Rodney Brooks · Factual · 19 Mar 2019 AI capabilities
Not checkable Ray Dalio Not checkable 18 Mar 2019 If you put your goals in the hands of RPs who can execute those goals well, and if you make it clear to them that they are personally… Ray Dalio · Not checkable · 18 Mar 2019
Not checkable Sam Altman Not checkable 18 Mar 2019 YC demo day and _Michael_ is the one wearing cargo shorts! Sam Altman · Not checkable · 18 Mar 2019
Not checkable Ray Dalio Not checkable 17 Mar 2019 That’s backward. Ray Dalio · Not checkable · 17 Mar 2019
Not checkable Ray Dalio Not checkable 16 Mar 2019 Ultimately, power will rule. Ray Dalio · Not checkable · 16 Mar 2019
Not checkable Ray Dalio Not checkable 16 Mar 2019 This is true of any system. Ray Dalio · Not checkable · 16 Mar 2019
Not checkable Ray Dalio Not checkable 15 Mar 2019 If people respect the rules of the idea meritocracy, there will be no conflict. Ray Dalio · Not checkable · 15 Mar 2019
Not checkable Ray Dalio Not checkable 14 Mar 2019 While all these principles exist for the well-being of the community, there may come times when adhering to them could threaten the… Ray Dalio · Not checkable · 14 Mar 2019
Not checkable Ray Dalio Not checkable 13 Mar 2019 I believe that what’s good must work well, and that having the organization work well is of paramount importance. Ray Dalio · Not checkable · 13 Mar 2019
Not checkable Ray Dalio Not checkable 12 Mar 2019 While all individuals have the right to have their own opinions, they do not have the right to render verdicts. Ray Dalio · Not checkable · 12 Mar 2019
Not checkable Ray Dalio Not checkable 12 Mar 2019 Part of the purpose of a believability-weighted system is to remove emotion from decision making. Ray Dalio · Not checkable · 12 Mar 2019
Not checkable Ray Dalio Not checkable 11 Mar 2019 In an idea meritocracy, there is bound to be more disagreement than in a typical organization Ray Dalio · Not checkable · 11 Mar 2019
Not checkable Ray Dalio Not checkable 9 Mar 2019 you can see this happening all the time in companies, organizations, and even political systems and nations. Ray Dalio · Not checkable · 9 Mar 2019
Mostly true Ray Dalio Factual 8 Mar 2019 Money donated to Grameen America is repeatedly lent and repaid, so each $1 donated results in roughly $12 lent to people over 10 years. Ray Dalio · Factual · 8 Mar 2019 Other
Not checkable Ray Dalio Not checkable 8 Mar 2019 By practicing open-mindedness and assertiveness, you should be able to resolve most disagreements. Ray Dalio · Not checkable · 8 Mar 2019
Unresolvable Ray Dalio Prediction 7 Mar 2019 2020 will be "the year of the ocean." Ray Dalio · Prediction · 7 Mar 2019 Other
Not checkable Ray Dalio Not checkable 7 Mar 2019 Almost every group that agrees on the big things ends up fighting about less important things and becoming enemies even though they should… Ray Dalio · Not checkable · 7 Mar 2019
Back to the list

Factual AI compute, hardware and data

“this push to make everything learnable is pushing the cost of AI outside that of individuals or even large university departments.”

Rodney Brooks, blog author 19 March 2019 A Better Lesson (rodneybrooks.com)

The push to make everything learnable is raising the cost of AI beyond what individuals or even large university departments can afford.

Mostly true True in substance, wrong in a detail.

The second judge would not resolve this claim; the verdict shown is the first judge's.

What would count as true or false

Judged as of March 2019. True if contemporaneous evidence shows that the compute and data costs of state-of-the-art learning-centred AI research (for example, training costs of leading deep learning or reinforcement learning models such as AlphaGo Zero, OpenAI Five or large language models of 2018–19) had risen beyond what individual researchers or typical large university departments could fund, so that frontier results were coming mainly from well-resourced companies. Mostly true if costs had clearly risen sharply and were out of reach for individuals, but the evidence shows large university departments could still afford some frontier-level work, or the link to the push for learnability is only partly supported. Mixed if the evidence is evenly split: costs were rising for some kinds of frontier work while comparable results remained affordable to academia. Mostly false if costs had risen only modestly and most frontier learning results were still achievable within university budgets, so the claim overstates a real trend. False if the evidence shows that AI costs at the time were not beyond the reach of individuals or university departments.

Evidence

  1. OpenAI Research | Conclusion OpenAI · 16 May 2018
    “We’re releasing an analysis showing that since 2012, the amount of compute used in the largest AI training runs has been increasing exponentially with a 3.4-month doubling time”
  2. OpenAI Five OpenAI · 25 June 2018
    “running on 256 GPUs and 128,000 CPU cores—a larger-scale version of the system we built to play the much-simpler solo variant”
  3. The Staggering Cost of Training SOTA AI Models | Synced Synced · 27 June 2019
    “Some of the models are so big that even in MILA (Montreal Institute of Learning Algorithm) we can’t run them because we don’t have the infrastructure for that. Only a few companies can run these very big models they’re talking about.”
  4. The Staggering Cost of Training SOTA AI Models | Synced Synced · 27 June 2019
    “Google suggests researchers with tight budgets could pretrain a smaller BERT-Base model on a single preemptible Cloud TPU v2, which takes about two weeks with a cost of about US$500.”

Reasoning

E1 reports that since 2012, the compute used in the largest AI training runs had grown exponentially, doubling every 3.4 months. E2 describes OpenAI Five running on 256 GPUs and 128,000 CPU cores. E3, from June 2019, quotes a researcher saying even MILA (Montreal Institute of Learning Algorithms) lacked the infrastructure to run some very large models and that only a few companies could run them. E4, however, reports Google's suggestion that researchers with tight budgets could pretrain a smaller BERT-Base model for about US$500, so some work stayed affordable, and the evidence ties the cost rise to the push for learnability only indirectly.

Cite this claim

Rodney Brooks, “The push to make everything learnable is raising the cost of AI beyond what individuals or even large university departments can afford.” (said 19 March 2019). Verdict: mostly true. Reliability API, claim c_ad1de70f030a, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_ad1de70f030a

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