George Hotz: Tiny Corp, Twitter, AI Safety, Self-Driving, GPT, AGI & God | Lex Fridman Podcast #387

George Hotz: Tiny Corp, Twitter, AI Safety, Self-Driving, GPT, AGI & God | Lex Fridman Podcast #387

Lex Fridman PodcastJun 29, 20233h 8m

Lex Fridman (host), George Hotz (guest), Narrator

Possible endpoints for human civilization: wireheading, super-stimulus media, societal collapse without species extinctionNature of intelligence, consciousness, and computation in humans vs large language modelsAI safety, centralization vs open source, and political power around advanced AITinyCorp, TinyGrad, and the push for decentralized, hardware-agnostic ML compute (TinyBox, TinyRack)Self-driving cars and Comma.ai’s DriveGPT approach using simulators and RLSoftware engineering philosophy: refactoring, simplicity, legacy code (Twitter), and toolingFuture of relationships, AI companions, VR/mixed reality worlds, and belief in God

In this episode of Lex Fridman Podcast, featuring Lex Fridman and George Hotz, George Hotz: Tiny Corp, Twitter, AI Safety, Self-Driving, GPT, AGI & God | Lex Fridman Podcast #387 explores george Hotz on AI Freedom, Human Obsolescence, and Decentralized Compute Futures George Hotz and Lex Fridman explore how advanced AI may transform and possibly end current human civilization, emphasizing wireheading, super-stimulus media, and AI-accelerated social manipulation. Hotz argues that intelligence itself is dangerous, but centralized control of AI is far worse than open access, making open source and decentralized compute his core mission with TinyCorp and TinyGrad. They discuss why current large language models aren’t AGI, why reproduction and robustness matter more than raw intelligence, and how self-driving via Comma.ai is increasingly an RL-in-simulation problem. The conversation also dives into software engineering culture, Twitter’s codebase, VR, AI girlfriends, God, and what it means for humans to “win” in a world saturated with machine intelligence.

George Hotz on AI Freedom, Human Obsolescence, and Decentralized Compute Futures

George Hotz and Lex Fridman explore how advanced AI may transform and possibly end current human civilization, emphasizing wireheading, super-stimulus media, and AI-accelerated social manipulation. Hotz argues that intelligence itself is dangerous, but centralized control of AI is far worse than open access, making open source and decentralized compute his core mission with TinyCorp and TinyGrad. They discuss why current large language models aren’t AGI, why reproduction and robustness matter more than raw intelligence, and how self-driving via Comma.ai is increasingly an RL-in-simulation problem. The conversation also dives into software engineering culture, Twitter’s codebase, VR, AI girlfriends, God, and what it means for humans to “win” in a world saturated with machine intelligence.

Key Takeaways

Intelligence is dangerous, but centralized intelligence is worse than widely distributed intelligence.

Hotz argues that advanced AI will inevitably be misused, but concentrating powerful models in the hands of a few corporations or governments guarantees a “chicken farm” scenario where the many are dominated by the few. ...

Get the full analysis with uListen AI

AI risk is more about bad humans using tools than AIs “waking up.”

He dismisses classic sci‑fi “rogue AI” plots, focusing instead on realistic threats: state actors and corporations using AI for mass persuasion, PSYOPs, and hyper-targeted content (e. ...

Get the full analysis with uListen AI

Reproduction and robustness define real “life” more than raw intelligence.

Hotz notes that all biological life is fundamentally reproductive and robust in nature, while current silicon systems, including robots and fabs, cannot self-replicate. ...

Get the full analysis with uListen AI

LLMs are impressive but fundamentally mid-tier and not AGI-level intelligence.

He views GPT-style models as powerful compressors trained with cross-entropy loss that produce “YouTube comment raps” and junior-engineer-level code. ...

Get the full analysis with uListen AI

Decentralized compute is a strategic defense against AI monopolies and nationalization.

TinyCorp’s mission is to build a minimal, hardware-agnostic ML stack (TinyGrad) and high-end local boxes (TinyBox/TinyRack) so individuals and small labs can run large models without cloud dependence. ...

Get the full analysis with uListen AI

Simplicity, refactoring, and strong tests beat complexity and promotion-driven bloat.

Drawing on Twitter’s codebase, he criticizes incentive structures where engineers get promoted for new libraries rather than deletion and simplification. ...

Get the full analysis with uListen AI

Self-driving progress hinges on realistic simulators and RL tuned to driver happiness.

At Comma. ...

Get the full analysis with uListen AI

Notable Quotes

Intelligence is scary. When I’m in the woods, the scariest animal to meet is a human.

George Hotz

You’re not gonna get paper‑clipped if everybody has an AI. You only get paper‑clipped if one group controls the AI.

George Hotz

Of course we were created by God. It’s the most obvious thing. I create worlds—how am I supposed to believe no one created ours?

George Hotz

We’ve solved the problem of self-driving… years ago. The question now is how to get a model to output a human driving policy people never want to disengage from.

George Hotz

You have never refactored enough. Your code can get smaller. Your ideas can be more elegant.

George Hotz

Questions Answered in This Episode

If intelligence is inherently dangerous, what concrete mechanisms—beyond open sourcing—can ensure that decentralized AI actually benefits individuals rather than just creating many new small tyrannies?

George Hotz and Lex Fridman explore how advanced AI may transform and possibly end current human civilization, emphasizing wireheading, super-stimulus media, and AI-accelerated social manipulation. ...

Get the full analysis with uListen AI

How realistic is Hotz’s belief that human civilization can “reset” after an AI-driven societal collapse, and what would it practically take for a low-tech taboo society to avoid repeating our mistakes?

Get the full analysis with uListen AI

To what extent can we trust AI systems to serve as personal “firewalls” against propaganda and PSYOPs, and how do we prevent those very filters from becoming new forms of subtle control?

Get the full analysis with uListen AI

Is Hotz underestimating the degree to which LLMs and code-generating models can evolve from mid-tier tools to genuine creative and reasoning partners, or is there a real ceiling without richer interaction and embodiment?

Get the full analysis with uListen AI

How should large organizations balance the short-term business risk of stopping to refactor legacy systems (like Twitter’s) against the long-term strategic risk of being unable to move fast due to accumulated complexity?

Get the full analysis with uListen AI

Transcript Preview

Lex Fridman

What possible ideas do you have for the w- how human species ends?

George Hotz

Sure. So I think the most obvious way to me is wire heading. We end up amusing ourselves to death. We end up all staring at that infinite TikTok and forgetting to eat. Maybe, maybe it's even more benign than this. Maybe we all just stop reproducing. Now, to be fair, it's probably hard to get all of humanity.

Lex Fridman

Yeah. The interesting thing about humanity is the diversity in it.

George Hotz

Oh, yeah.

Lex Fridman

Organisms in general. There's a lot of weirdos out there.

George Hotz

Well-

Lex Fridman

Two of them are sitting here.

George Hotz

I mean, diversity in humanity is-

Lex Fridman

With due respect. (laughs)

George Hotz

(laughs) I wish I was more weird.

Lex Fridman

The following is a conversation with George Hotz, his third time on this podcast. He's the founder of Comma.ai that seeks to solve autonomous driving, and is the founder of a new company called TinyCorp that created TinyGrad, a neural network framework that is extremely simple, with the goal of making it run on any device by any human easily and efficiently. As you know, George also did a large number of fun and amazing things, from hacking the iPhone to recently joining Twitter for a bit as an "intern" in quotes, making the case for refactoring the Twitter code base. In general, he's a fascinating engineer and human being, and one of my favorite people to talk to. This is the Lex Fridman Podcast. To support it, please check out our sponsors in the description. And now, dear friends, here's George Hotz. You mentioned something in a stream about the philosophical nature of time. So, uh, let's start with a wild question. Do you think time is an illusion?

George Hotz

You know-

Lex Fridman

(laughs)

George Hotz

... I sell phone calls, uh, to Comma for $1,000. Uh, and some guy called me, and, uh, like, you know, it's $1,000, so you can talk to me for, for half an hour. And he's like, "Uh, yeah, okay. So like, time doesn't exist, and I really wanted to share this with you." I'm like, "Well, what do you mean time doesn't exist," right? Like, I think time is a useful model whether it exists or not, right? Like, does quantum physics exist? Well, it doesn't matter. I- i- it's about whether it's a useful model to describe reality. Is time maybe compressive?

Lex Fridman

Do you think there is an objective reality or is everything just useful models? Like underneath it all, is there an actual thing that we're constructing models for?

George Hotz

I don't know.

Lex Fridman

I was hoping you would know.

George Hotz

I don't think it matters.

Lex Fridman

I mean, this kind of connects to the models we construct of reality with machine learning, right?

George Hotz

Sure.

Lex Fridman

Like, is it just nice to have useful approximations of the world such that we can do something with it?

George Hotz

So there are things that are real. Kolmogorov complexity is real.

Install uListen to search the full transcript and get AI-powered insights

Get Full Transcript

Get more from every podcast

AI summaries, searchable transcripts, and fact-checking. Free forever.

Add to Chrome