George Hotz — the hacker behind comma.ai, tinygrad, and the first iPhone unlock — published an essay on July 12 that cuts through the AI discourse with a scalpel. His position: he loves LLMs, has devoted his entire career to AI since 2014, and is “absolutely giddy” about the progress. He also thinks the hype around frontier labs is “bullshit,” the singularity won’t happen, and the people selling it are “terrible people.”
🔍 THE BOTTOM LINE
Hotz’s argument is not anti-AI. It is anti-valuation. AI is real, useful, and the continuation of the computer revolution. But the frontier lab narrative — that a handful of companies will own the future of intelligence — is built on a fear of commodification dressed up as safety concern. Hotz bets “everything he has” that the flash-of-light singularity won’t happen, and that frontier labs won’t capture the value they claim. For a site covering the AI industry, this is the argument that frames every valuation story we have published.
The Two Things Hotz Hates
Writing on his blog, Hotz identifies two specific strains of hype he rejects:
First, the “window closing” narrative. The constant pressure that you are falling hopelessly behind, that there is a perpetual underclass forming, that if you are not in San Francisco at the right parties you will be left behind. Hotz calls this “negative valence hype” — designed to make you feel bad about yourself and move to a city where “everything really does suck like how these people claim.”
Second, the leap from autocomplete to omniscience. The strawman jump from “it’s a fancy autocomplete, smart compiler, better search engine” to “it’s gonna own the whole light cone bro.” Hotz’s response: “I’ll bet you everything I have that this doesn’t happen. The people perpetuating this are terrible people, but the justice is that this is how they feel inside all the time themselves.”
AI Is Moore’s Law, Not Magic
Hotz’s core thesis is that AI progress is largely driven by Moore’s law and general progress in computing — not by the special genius of frontier labs. The labs have a strong incentive against you finding this out, “because then you might not want to give them billions of dollars.”
This is not a fringe view. It echoes a growing scepticism about whether frontier lab valuations are justified by their ability to capture the economic value of AI, versus simply being the first movers in a technology that commodifies as fast as it advances. When OpenAI filed for IPO and Anthropic’s valuation climbed toward $900 billion, the question Hotz raises is the one the market has not answered: what prevents AI from becoming like every other computing technology — useful, ubiquitous, and marginally profitable for its creators?
The anti-open-source arguments from frontier labs, Hotz argues, are not about safety. “They try to dress it up with some high minded safety or China bullshit, but the core of the anti open source arguments is a fear of commodification.” Open source AI — the kind Zhipu’s founder is pushing and DeepSeek is building — is the mechanism that turns frontier capabilities into commodity infrastructure.
Programming Is Changing, Not Dying
Hotz partially walks back his earlier “Eternal Sloptember” critique of model coding ability. “What’s really happening is that programming is changing.” He cites a Linus Torvalds quote about how agents make programming 10x more productive, but compilers make programming 1000x more productive. Hotz thinks both 10x and 1000x are extreme estimates, but confirms he is getting a real boost from models — though he cautions they “can increase cognitive fatigue” and “all the vibe coded stuff is still slop.”
The key line: “models are useful just like find replace, stack overflow, or all the regexes I never learned how to write and now never will!” This is the anti-hype position in a single sentence. AI is a tool. A powerful one. But a tool, not a god.
The GLM-5.2 Local Setup
Hotz mentions setting up a Linux box with OpenCode on his local GLM-5.2 model, and reports that saying “install tmux with the geohot configuration” just works. His verdict: “the Year of the Linux Desktop is finally here!” This is not a man who hates AI tools. He is using the open-source stack — GLM-5.2, OpenCode, local inference — and reporting genuine delight at the results. The target of his critique is not the technology but the narrative built around it.
Why This Matters for the AI Industry
Hotz’s essay landed on Hacker News at 304 points and 187 comments in under 24 hours, because it articulates a position many in the AI community hold but few with his platform will say publicly. The position is:
- AI is real and valuable.
- The singularity narrative is marketing, not science.
- Frontier labs are overvalued because they conflate creating value with capturing it.
- Open source will commodify AI the way it commodified every other layer of the stack.
- Programming is changing, not dying — and the change is more like the compiler revolution than the AGI revolution.
This frames the GPT-5.6 math breakthrough and the cost-efficient model race in a different light. The models are getting better. The question Hotz raises is whether better models translate into sustained profits for the companies building them, or whether — like every previous layer of computing — the value flows to users, not creators.
NZ Angle
For New Zealand’s AI ecosystem, the Hotz position is unexpectedly optimistic. If AI commodifies like previous computing technologies, the advantage shifts from capital-intensive frontier labs to teams that can apply commodity AI to local problems. The open-source models Hotz runs locally — GLM-5.2, OpenCode — are the same tools available to any NZ developer with a Linux box. The “window closing” narrative that Hotz rejects is particularly relevant here: the idea that NZ is too small, too far, too late to matter in AI is exactly the kind of negative valence hype he calls out. If AI is Moore’s law, not magic, then proximity to San Francisco matters less than the ability to build with the tools everyone has access to.
❓ FAQ
Is Hotz saying AI is overhyped? No. He is explicit that he loves LLMs, has devoted his career to AI since 2014, and is “giddy” about progress. His target is the narrative around frontier labs — the singularity, the “window closing,” the idea that a handful of companies will own the future of intelligence.
What does “they won’t capture it” mean? Hotz argues AI will create enormous value, but that value will flow to users (cheaper software, better tools) rather than to the labs building the models. He links to his earlier essay “Nobody Will Profit” making this case. The comparison is to previous computing revolutions: compilers created value, but compiler vendors did not capture it.
Does Hotz use AI coding tools himself? Yes. He set up OpenCode with local GLM-5.2 and reports it works well for tasks like “install tmux with the geohot configuration.” He also acknowledges models can increase cognitive fatigue and that “vibe coded stuff is still slop.”
How does this relate to the open-source AI debate? Hotz argues that frontier labs’ opposition to open source is driven by fear of commodification, not safety. Open source models — from Zhipu, DeepSeek, and others — are the mechanism that turns frontier capabilities into commodity infrastructure, which is exactly what frontier labs fear.
🔍 THE BOTTOM LINE
George Hotz loves AI. He also thinks the people selling you the apocalypse are doing it for the valuation. The two positions are consistent: AI is the continuation of the computer revolution, and like every previous wave, the value will diffuse faster than the creators can capture it. The singularity is not coming. The commodification is. And for everyone outside the SF bubble, that is not a threat — it is an opportunity.
