geohot Criticizes LLM Hype and Exaggerations

Renowned hacker and founder of AI engineering company Comma.ai, George Hotz (known as geohot), has expressed his love for large language models (LLMs) and his strong disapproval of the hype surrounding them in a recent blog post. Reported by therepanic on Hacker News (Best), the post has sparked significant discussion within the industry.

Hotz begins by clarifying, “I don’t want anyone reading this blog to misunderstand how much I love AI,” emphasizing his deep passion for the field and his career-long dedication to artificial intelligence. He recounts his journey, which began with hacking from 2007 to 2014, after which he fully committed to AI. Hotz acknowledges the genuine promise of advancements in LLMs, self-driving cars, generative video models, and coding agents.

Criticism of Two Types of Hype

Hotz’s critique specifically targets two types of hype.

The first is what he calls “negative valence hype.” He refers to statements like “The AI window is closing,” “You’ll be left behind in a permanent underclass,” or “We’re falling hopelessly behind.” He argues that these claims are not only false but also create unnecessary panic, leading people to migrate to overcrowded and challenging environments like San Francisco.

The second is “leaps of logic.” Hotz ridicules arguments that jump from the current state of AI as advanced autocomplete, smart compilers, or excellent search engines to apocalyptic visions like “AI will control the universe” or “One day, a flash in the sky will change everything.” He confidently asserts, “I’d bet everything I own that such a future will not come to pass.”

Hotz criticizes those who spread such exaggerated claims, describing them as “terrible people.” He believes their own internal anxieties drive them to project similar fears onto others.

Changing Nature of Programming and New Skills

Hotz reflects on a previous blog post, “The Eternal Sloptember,” where he criticized the programming capabilities of current models. He admits he might have been overly harsh but points out that “what’s happening now is a transformation in programming itself.”

Quoting Linus Torvalds, who said that compilers increased programming productivity by a factor of 1,000, Hotz evaluates AI agents as tools that can enhance productivity tenfold—while noting that these estimates might be overly optimistic.

He emphasizes that effectively leveraging AI requires new skills and careful attention. Hotz warns of the potential for “increased cognitive fatigue” and firmly states, “Code written with just vibes is still sloppy.” Despite the promise of improved productivity, he questions, “Where is the new magical software?”

Nevertheless, Hotz acknowledges that AI models are becoming useful tools, comparable to “search-and-replace functions, Stack Overflow, or regex patterns I never learned to write.”

Criticism of Frontier Labs and “Enclosure”

At the heart of Hotz’s critique lies his skepticism toward the valuation of frontier labs. “My main argument against the valuation of frontier labs is not that AI doesn’t generate value, but that they can’t capture that value,” he explains.

Hotz argues that AI advancements result not from the creation of specific labs but from the overall progress of Moore’s Law and general computational advancements. He asserts that opposition to open source is rooted in “a fear of commoditization.” Claims of prioritizing safety or concerns about competition from China, he suggests, are ultimately about protecting the value of proprietary technologies.

While Hotz acknowledges that he “may have been too harsh,” his candid analysis exposes the true motives behind these arguments and provides a sharp critique of the current state of the tech industry. This perspective aligns with broader calls for transparency in the tech world, as seen in recent investigative reporting that helped identify the Scattered Spider suspects and shed light on Windows GDID issues.

Editorial Opinion

In the short term, this post will likely serve as a wake-up call for the AI industry, highlighting the gap between AI hype and its practical applications. As funding rounds become increasingly competitive, investors may reconsider the valuation of frontier labs. Within the developer community, there could be a growing recognition that “AI is not omnipotent,” leading to its positioning as a more realistic and functional tool.

In the long term, if the “commoditization” of AI that Hotz predicts does occur, it could fundamentally reshape the current business models of the AI industry. Just as with Moore’s Law, the inability of a few companies to monopolize value might accelerate the rise of open-source models while hastening the decline of closed frontier models.

The editorial team is particularly intrigued by Hotz’s point about the “cognitive fatigue” caused by AI. The issue of increased cognitive load on developers, despite promises of enhanced productivity, remains underexplored. How this side effect will impact the adoption of AI tools is a matter for further investigation.

References

• “I love LLMs, I hate hype”, by therepanic — Hacker News (Best), 2026-07-12T18:31:56.000Z (ARR)
• Source URL: https://geohot.github.io//blog/jekyll/update/2026/07/12/i-love-llms.html