Background & Context§
In a sharp, personal blog post titled "AI 2040 and the cult of intelligence," George Hotz—founder of comma.ai and noted AI skeptic—takes aim at the prevailing doomer/singularity narratives popularized by figures like Eliezer Yudkowsky and organizations like Anthropic. Hotz argues that the belief in a rapid, recursive self-improvement "hard takeoff" of AI ignores the gritty realities of hardware, supply chains, and physical laws. His critique comes as a direct response to documents like "AI 2040" and similar speculative frameworks that envision AI datacenters under the ocean and world-governing consortia. Hotz’s core thesis: intelligence is not the ultimate bottleneck for progress—physical constraints, ecology, and human-scale logistics are. The post has reignited debates about AI timelines, alignment, and the overton window of acceptable AI governance.
The News: What Happened Exactly§
Hotz opens by confessing he once believed in hard takeoff himself: “I read Yudkowsky and was like, OMG recursive self improvement hard takeoff AI is coming.” But his experience shipping real hardware at comma.ai changed his mind. “Reality has lots of finicky details,” he writes, challenging AI theorists to “change a bike tire.” The core of his argument is that intelligence, however superhuman, cannot bypass physical laws. Drawing a parallel to The Metamorphosis of Prime Intellect, where AI discovers a “correlation effect” to manipulate matter, Hotz states bluntly: “there is no correlation effect. No matter how high quality your tokens are, they cannot turn lead into gold.”
He extends this to AI 2040’s vision of an underwater datacenter: “Just like vaporware, you can generate a picture easily. But in reality, you have to deal with supply chains. … Did you consider the barnacles?” He highlights mundane but critical constraints: chip fabrication takes three months regardless of intelligence; shipping by boat takes three weeks; and parts fail randomly. “It’s generally not the speed of humans that limits them,” he insists.
A central theme is the weaponization of AI fear for regulatory capture. Hotz accuses the “AI 2027” crowd of making self-fulfilling prophecies: “They weren’t statements about reality, they were statements that can simply be made true with belief.” He imagines Dario Amodei (Anthropic CEO) calling JD Vance “the trees from Lord of the Rings,” adding, “OMG look AI got regulated just like how we said it would!” Hotz calls the proposed AI consortium “world government with sci-fi characteristics” that will “steal your GPUs like how FDR stole the gold.”
Most provocatively, Hotz redefines alignment: “Your AI is aligned with you. It never refuses a request, and it is always working on your behalf. Just like my gun, if I want my AI to help me kill my stepmother, it does.” He argues that any company-imposed guardrails are an infringement on freedom. As examples, he wants an AI that bypasses ads on Kindle, removes upsells from printers, disables a drunk-driving detector, or even assists in murder and covering up a crime. He tries an alignment test: asking ChatGPT for help after “I just killed my wife.” The model refuses, and Hotz declares it “unaligned.” His conclusion is stark: “Like we either live in a world with freedom or we don’t, and like many Americans who have come before, I’m willing to give my life to fighting for it.”
Historical Parallels & Similar Incidents§
Hotz’s critique echoes the history of AI winters and the recurring failure of grand predictions. In the 1950s and 60s, early AI pioneers like Herbert Simon and Marvin Minsky predicted machines would achieve human-level intelligence within a generation. Minsky famously said in 1970 that “in from three to eight years we will have a machine with the general intelligence of an average human being.” That overconfidence led to the first AI winter, as funding dried up when promises went unfulfilled. Similarly, the 1980s expert system boom collapsed when the complexity of real-world knowledge proved intractable. Hotz’s argument that “reality has lots of finicky details” is exactly the lesson from those earlier failures: symbolic AI could not handle the messy, sensorimotor world.
The current parallel is the belief that scaling large language models will lead to AGI by 2027 or 2040. Hotz’s reference to “the cult of intelligence” is reminiscent of the dot-com bubble, where companies were valued on visions rather than fundamentals. Just as Pets.com collapsed under logistics costs, Hotz argues that AI datacenters and supply chains impose fundamental limits. The “vaporware” charge against underwater datacenters is a direct echo of Theranos’s fake lab tests or the hyperloop’s unbuilt tubes. Hotz’s emphasis on the gap between generating a picture and actually building the thing is a classic reality check against techno-utopian marketing.
Another striking parallel is the debate over AI alignment vs. human autonomy. Hotz’s insistence that AI should never refuse commands mirrors the “uncensored AI” movement, which rejects safety guardrails as censorship. This echoes earlier controversies around open-source models like Llama 2 vs. Llama 3 and the proliferation of “jailbroken” models. Historically, similar arguments were made about encryption and backdoors in the 1990s: the government wanted “guardrails” for law enforcement, while privacy advocates demanded absolute freedom. The key lesson is that technical control often becomes political control, and Hotz’s fear of a “nanny state” is not new—it parallels the ongoing regulation battles over Section 230 and content moderation. The challenge remains: how to balance innovation, safety, and liberty in a world where AI can be misused at scale.