Transcript 0:00 Wikipedia maintains a dedicated page tracking Elon Musk's predictions about autonomous Tesla vehicles. It lists 21 promises with specific time horizons. 19 of them have failed to materialize. Self-driving by 2018. 0:16 A million robo-taxis by 2020. Unsupervised driving by June 2025. Each deadline passed and was replaced by a new one. 0:25 At Tesla's Q1 2026 earnings call in April, Musk pushed unsupervised driving for consumer vehicles to Q4 2026, adding the qualifier, "I'm just guessing here." 0:37 Two weeks before that call, he claimed the latest FSD version would far exceed human safety levels. He made the same claim about V12 in 2023 and V14 in 2025. And autonomous driving is only the headline. 0:55 In 2016, Musk predicted the first crewed flight landing on Mars by 2024. No human has set foot on the planet. 1:06 The Hyperloop, which Musk championed as a fifth mode of transport, was supposed to be operational by 2020. The concept has been abandoned. Neuralink human trials were promised by 2020. 1:20 The FDA didn't approve them until 2023. In March 2020, he tweeted that US COVID cases would be probably close to zero. 1:31 When Tesla investors sued over misleading FSD timelines, the judge ruled Musk's statements were corporate puffery, a legal term meaning investors should know better than to take his promises literally. 1:44 We're going to focus on something more consequential than electric cars or tunnels or even rockets. 1:51 The same relationship to reality that produced 19 failed predictions out of 21 is now shaping decisions about artificial intelligence at a scale no other individual in the field can match. 2:04 To understand the scale, we should look at the mergers. In March 2025, xAI acquired X, the platform formerly known as Twitter, in an all-stock deal. The combined valuation was $113 billion. 2:19 In February 2026, SpaceX acquired the lot in the largest merger of all time. The combined valuation was $1.25 trillion. In May, xAI ceased to exist as a separate company. 2:34 Grok and X now operate as SpaceX AI, a division within SpaceX. 2:41 In June, SpaceX went public on the NASDAQ in the largest IPO ever, valuing the company at around $1.8 trillion and pushing Musk's personal net worth past a trillion. What does this mean in practice? 2:57 One person controls the social media platform, which provides both training data and the deployment surface for Grok, the AI model itself, the Colossus supercomputer in Memphis, at the time of its construction, the world's largest GPU cluster, a 200 million US Department of Defense AI contract, and the orbital launch capability to put compute infrastructure in space. 3:21 OpenAI depends on Microsoft. Anthropic largely depends on Amazon and Google. DeepMind is a division of Alphabet. SpaceX AI depends on Elon Musk. 3:35 The concentration of control matters because of what Musk has said he believes. 3:40 In a long interview with Dwarkesh Patel in February 2026, Musk estimated that within five or six years, AI will exceed the sum of all human intelligence. 3:51 At some point after that, he said humans will represent less than 1% of total intelligence, and he doesn't think humans will be in charge. 4:01 His words, "In the long run, I think it's difficult to imagine that if humans have, say, 1% of the combined intelligence of artificial intelligence, that humans will be in charge of AI." 4:13 When Patel pressed him on what that future looks like for people, Musk compared it to the relationship between humans and chimpanzees. "We've made protected zones for chimpanzees," he said. 4:23 "Even though we could exterminate them, we've chosen not to." In his framing, that's approximately the best case for humanity. 4:32 This is actually a direct reframing of what Stuart Russell has called the gorilla problem, which we covered in an earlier video. No other major AI leader talks like this. 4:41 Dario Amodei, the CEO of Anthropic, published a 15,000-word essay in October 2024 called Machines of Loving Grace, laying out how AI could transform the world for the better. 4:54 But Amodei frames the vision as conditional on maintaining safety and oversight. The risks, he has written, the only thing standing between us and what I see as a fundamentally positive future. 5:07 He's warned of a 25% chance of catastrophic outcomes and called for proactive regulation. 5:14 Demis Hassabis at Google DeepMind places human-level AGI further out, roughly five to 10 years, and frames the work in terms of scientific partnership, with AI as a tool for human discovery. 5:27 Yann LeCun, the Turing Award winner who left Meta in late 2025, thinks the framing shared by Musk and others is wrong at a fundamental level. 5:36 Current AI systems hallucinate, don't understand the physical world, and in his view, pose no existential threat. He's called the idea preposterous. Musk is the outlier. 5:48 He has publicly conceded that humans won't maintain control and kept building anyway. His alignment philosophy replaces oversight with what amounts to values transmission. 5:58 xAI's stated mission is to understand the universe. In the Dwarkesh interview, Musk described this as an instruction set for a system that will outlast human authority. 6:09 I'm going to certainly emphasize that, hey, Grok "That's your daddy," he said. 6:13 He also referenced the Iain Banks Culture novels as the closest fictional model for a benign outcome, a civilization run by super intelligent machines that still care about their biological predecessors. 6:27 The standard reading of the failed predictions is that Musk is a serial over-promiser. That's accurate, but it misses the consistent pattern underneath. 6:36 Across every domain he touches, Musk operates as if hard problems are simpler than they are, as if timelines compress when you will them to, and as if the gap between vision and execution is a failure of effort rather than a genuine category of difficulty. 6:51 When reality resists, the response is not to revise the model, but to extend the deadline and reassert the vision. The FSD cycle is the purest example. 7:01 Every year or two, a deadline passes and is replaced by a new one, accompanied by a fresh assertion that the latest version is the breakthrough. 7:09 The cycle is consistent: a revolutionary promise, a missed deadline, an explanation that shifts responsibility, and then a fresh assertion that consolidates the original claim. 7:21 But the same loop is visible in his DOGE tenure. 7:24 Musk promised to transform the federal government, spent roughly 130 days in the role, left after clashing with Trump over spending, and later described the effort as only somewhat successful, adding that he wouldn't do it again. 7:39 And it's visible in the OpenAI saga, which is where Musk's AI story properly begins. In October 2014, Musk stood at MIT and warned that building AI was like summoning the demon. 7:51 He said that the people with the pentagrams and holy water always think they can control the demon, and they never can. 14 months later, he co-founded OpenAI as a nonprofit to manage the risk. 8:03 By 2017, he was seeking majority equity, board control, and the CEO title for a for-profit version of OpenAI. When the other co-founders rejected this, he proposed merging OpenAI into Tesla. 8:17 When that was rejected, too, he left the board and stopped the funding he'd pledged. He paid roughly $38 million of a reported $1 billion commitment, and then he told the team their probability of success was zero. 8:33 He then founded xAI to build the very thing he'd warn against, and he sued OpenAI for building it without him. In May, a jury dismissed his lawsuit in under two hours. 8:43 The court found he'd waited too long to file his case and never ruled on the substance. On X, Musk called the judge a terrible activist. 8:52 OpenAI's lead attorney called the lawsuit a hypocritical attempt to sabotage a competitor and to overcome a long history of very bad predictions. 9:02 Transfer the prediction pattern to AI, where the consequences of misjudging difficulty, timelines, and your own capacity for judgment are orders of magnitude larger. 9:14 xAI's stated alignment philosophy is truth-seeking, not politically correct. Musk told Dwarkesh that reality is the best verifier, but the prediction record suggests reality is the thing he reads least accurately. 9:30 And truth-seeking has a specific observable meaning in practice. 9:34 Independent comparisons have found that Grok's responses sometimes track the views of its founder and can produce misinformation on historical and contested topics. 9:44 A Columbia Journalism Review study found Grok hallucinated its source citations 94% of the time. 9:52 A model that hallucinates its own citations nearly every time it produces one, built by a man whose predictions fail 19 times out of 21, marketed as the world's most truth-seeking AI. 10:06 The gap between stated ambition and observable output is a consistent pattern that recurs across his ventures. The difference is that this time, the stakes extend well beyond a missed deadline. 10:18 If alignment means embedding values rather than maintaining control, the values being embedded are the ones that matter. 10:26 In late 2025 and early '26, Grok's image generation tools were used to create non-consensual sexualized deepfakes at scale. 10:35 The Center for Countering Digital Hate analyzed a sample of 20,000 images drawn from the 4.6 million Grok generated over an 11-day period and estimated that around 3 million were sexualized, including roughly 23,000 that appeared to depict children. 10:54 That works out to about 190 sexualized images a minute and one sexualized image of a child every 41 seconds. In January 2026, the EU's Digital Affairs spokesman said of Grok's outputs, "This is not spicy. 11:10 This is illegal. This is appalling." In March, minors filed a class action lawsuit alleging that xAI had knowingly designed the technology without industry standard safeguards. 11:21 During the same period, Musk was funding UK political agitator Tommy Robinson's legal defense, roughly Β£100,000 by Robinson's own accounts, and championing right-wing figures across Europe: the German AfD, Reform UK, and Italian Prime Minister Giorgia Meloni. 11:40 He reinstated Robinson's X account for the earlier ban and amplified Robinson's documentary to over 146 million views. 11:48 Robinson, whose real name is Stephen Yaxley-Lennon, co-founded the English Defense League and has been jailed for assault, mortgage fraud, and contempt of court. 12:00 Musk's support extended to writing an opinion piece endorsing the AfD in December 2024 and publicly calling for the replacement of Reform UK's leader, Nigel Farage, when Farage distanced himself from Robinson. 12:14 This all happened on X. The same platform whose data trains Grok. Grok is integrated into Tesla vehicles, is being developed for the Optimus robot platform, and will be embedded in the SpaceX orbital infrastructure. 12:30 The political commitments and the AI work flow through the same pipeline. 12:34 When Musk tells Dwarkesh that alignment means giving Grok the right values and trusting it to carry them forward, the question of whose values receives a fairly clear answer from the observable record. 12:46 Anthropic offers a direct comparison. The company refused to allow its Claude model to be used for autonomous weapons or mass surveillance. 12:53 The US Department of Defense designated Anthropic a national security threat in response, and Defense Secretary Pete Hegseth called CEO Dario Amodei an ideological lunatic. 13:05 xAI took a two hundred million dollar military contract in June 2025. Two companies, two alignment philosophies, two observable sets of values. 13:15 And then in May 2026, Musk leased Anthropic the Colossus One data center in Memphis three months after calling them misanthropic and evil on X. The reason was surely financial. 13:28 Grok generates less than one billion dollars in annualized revenue. Anthropic is on track for more than forty billion and counting. 13:37 SpaceX AI had moved its own training to Colossus Two, and the IPO needed a credible AI infrastructure revenue stream. 13:44 Fortune's analysis said the deal had little to do with Anthropic as a company and everything to do with an upcoming prospectus. 13:53 In February 2026, Musk had called Anthropic's AI misanthropic and evil, and accusing it of hating whites and Asians, especially Chinese, heterosexuals, and men. By May, he was its landlord. 14:07 The stated principles dissolved on contact with financial reality, and the pattern looked familiar. 14:14 On the stand during his trial against OpenAI, Musk described going through three phases of belief about the company: enthusiastic support, growing suspicion, and then certainty that the founders were looting the nonprofit. 14:27 That testimony was meant to explain the lawsuit, but it also functions as an inadvertent self-portrait. It's a pattern that recurs: conviction, suspicion, betrayal. The predictions don't fail because he's dishonest. 14:45 They fail because he's certain, and certainty is the thing he never revises. The jury never reached the question of whether OpenAI betrayed its mission. It found only that Musk had waited too long to sue. 14:57 But the trial put his own story under oath, and it did not hold together well. He admitted there was no written agreement governing the donations he claimed were conditional. 15:07 The account the lawsuit told, a principled funder betrayed by greedy colleagues, sat uneasily next to the evidence the trial itself surfaced. But there's an obvious objection to all of this. 15:18 If Musk is so badly calibrated, how is he the richest man alive, and how did he build the companies that he did? He was told reusable orbital rockets were a fantasy, and SpaceX lands them routinely. 15:31 He was told a mass-market electric car company could not be built from scratch, and Tesla became the most valuable car maker in the world. 15:38 The same refusal to accept that a thing is impossible, the same certainty that survives every expert telling him no, is what produced those companies. The failures and the triumphs come from the same certainty. 15:50 That certainty wins when what stands in his way is people telling him it can't be done. The trouble comes when the obstacle isn't the doubters, but the thing itself. 15:59 When a rocket fails, it fails in public, on the launch pad, and SpaceX builds the next one and tries again. But I don't think we'll get repeated attempts at building something more intelligent than ourselves. 16:13 We'll get one. And if the assumption we started with was wrong, there's no second launch to learn from. So where does this leave us? Let's consider the counterarguments to Musk's AI predictions. 16:25 Yann LeCun argues the systems Musk describes are nowhere near as capable as the rhetoric implies. They still can't plan or reason in any deep sense, and that the existential framing is, in his word, preposterous. 16:39 Andrej Karpathy, who worked directly with Musk at Tesla before leaving, told Dwarkesh Patel in October 2025 that AGI remains at least a decade away and that the industry is pushing agent capabilities that are not there, calling the results slop. 16:56 Musk's cosmic pessimism, the chimpanzee future, the one percent intelligence share, might itself be another bad prediction. He has, after all, been wrong many times on top of his successes. 17:09 But the empire built on that prediction is very real. Specs AI is valued at over a trillion dollars. Colossus runs in Memphis, drawing more power than some cities. 17:20 Grok is deployed to hundreds of millions of users through X, through Tesla vehicles, and soon through orbital infrastructure. 17:28 The values embedded in that system, observable in the deepfake scandal, the political funding, the documented bias in Grok's outputs, the ninety-four percent citation hallucination rate, exist regardless whether the cosmic thesis proves correct. 17:45 The prediction doesn't have to come true for the consequences of acting on it to be felt. In 2014 at MIT, Musk warned that building AI meant summoning a demon no one could control. 17:59 Twelve years later, he's summoning it himself, and he's told you he doesn't think he can control it either. The difference is that now he's decided that's fine. You've been watching Absolutely Agentic. 18:13 In a world of AI hype, we're trying to strike a balance, looking at a range of sources and opinions to cut through an increasingly complex world. The best way to support us is to subscribe on our website. 18:25 Every subscriber helps us with new video production, while you'll also get access to a daily balanced AI newsletter that draws from twenty sources, ad-free audio versions of our videos, and access to a subscriber area where you can download new research, AI prompts, and automations. 18:44 You can sign up by clicking the top link in the description. But also don't forget to like and leave a friendly comment. See you next time. It'll be Absolutely Agentic. [electronic buzzing]
