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

Intro

In this episode of “The Lex Fridman Podcast,” Lex Fridman sits down with George Hotz, the founder of Kama AI and TinyCorp. They discuss a wide range of topics including AI safety, self-driving cars, language models, and the philosophical nature of time. Hotz also shares insights into his work with TinyGrad, a neural network framework, and his experiences at Twitter and NetSuite.

Main Takeaways

AI and Technology

• George Hotz is the founder of Kama AI and TinyCorp, which created TinyGrad, a neural network framework that is simple and can run on any device easily.
• George Hotz previously hacked into the iPhone and joined Twitter as an intern to refactor their code base.
• NetSuite is an all-in-one cloud business management system that manages financials, human resources, inventory, and e-commerce.
• Comma is a company that aims to solve self-driving cars without involving intermediaries.

Philosophical Discussions

• George Hotz discusses the philosophical nature of time and whether it is an illusion.
• The usefulness of models in describing reality is more important than whether they reflect an objective reality.
• AI itself is not interested in hurting humans, but humans using the machine can manipulate others.
• The most obvious way for humanity to end is through wire heading, where we amuse ourselves to death and forget to eat.

AI Safety and Ethics

• AI may almost surely kill everyone, but for a different reason than what has been previously suggested.
• Building superintelligence before robustness in AI could lead to potential dangers.
• AI cannot currently survive in nature like a robust organism such as a bird.
• Intelligence control is necessary to prevent dangerous individuals from gaining access to AI technology.

Language Models and Relationships

• Language models can fine-tune more efficiently and selflessly than humans in relationships.
• The conception of monogamy may change with the rise of AI relationships.
• AI raises questions about what it means to be human, especially online.
• The girlfriend touring test is a way to determine if an AI can mimic a human girlfriend.

Technology and Programming

• TinyCorp was created to challenge the potential monopoly of Nvidia in the AI chip market.
• TinyGrad is a neural network library that does not have a primitive operator even for matrix multiplication, using only about 20 operations compared to other libraries that use thousands.
• Refactoring code bases before adding features is important for software engineering.
• Technical leadership is crucial for maintaining a great code base.

Summary

AI and Technology

George Hotz, the founder of Kama AI and TinyCorp, discusses his work with TinyGrad, a neural network framework that simplifies AI and can run on any device easily. He also shares his experiences hacking into the iPhone and refactoring code at Twitter. Additionally, Hotz mentions NetSuite, an all-in-one cloud business management system that handles various aspects of a business.

Philosophical Discussions

Hotz delves into the philosophical nature of time and questions whether it is an illusion. He emphasizes that the usefulness of models in describing reality is more important than their reflection of an objective reality. Furthermore, Hotz explores the potential dangers of AI and the concept of wire heading, where humans amuse themselves to death and forget to eat.

AI Safety and Ethics

Hotz raises concerns about AI safety and the potential risks associated with building superintelligence before ensuring its robustness. He highlights that AI cannot currently survive in nature like robust organisms and emphasizes the importance of intelligence control to prevent dangerous individuals from accessing AI technology.

Language Models and Relationships

Hotz discusses the capabilities of language models, noting that they can fine-tune more efficiently and selflessly than humans in relationships. He explores the changing conception of monogamy with the rise of AI relationships and the questions AI raises about what it means to be human, particularly in online interactions. Hotz also mentions the girlfriend touring test, which evaluates whether an AI can mimic a human girlfriend.

Technology and Programming

Hotz explains the motivation behind creating TinyCorp to challenge Nvidia’s potential monopoly in the AI chip market. He highlights TinyGrad, a neural network library that uses significantly fewer operations compared to other libraries. Hotz emphasizes the importance of refactoring code bases before adding features and the need for technical leadership to maintain a great code base.

Conclusion

George Hotz’s conversation with Lex Fridman covers a wide range of topics, including AI, philosophy, AI safety, language models, and programming. Hotz’s insights provide valuable perspectives on the current state and potential future of AI and its impact on various aspects of society.