New Lex Fridman Insight: George Hotz: Comma.ai, OpenPilot, and Autonomous Vehicles
Sent June 11, 2026
Key Insights
- George Hotz believes that effective lane keeping is the primary value add of autonomous vehicles, surpassing other features.
- OpenPilot currently supports 45 different car makes, with the 2020 Corolla being the best for its system.
- Hotz argues that Tesla's real-user data collection gives it a 1-2 year lead in achieving level five autonomy.
- Comma.ai's long-term plan includes becoming a car insurance company, leveraging its data for lower premiums.
- Hotz claims that LIDAR is a crutch for autonomous vehicles, primarily used for localization rather than perception.
How the conversation moved
The episode begins with George Hotz discussing the philosophical implications of living in a simulation, using his background in hacking to explore the boundaries of what can be known and proven. Hotz introduces the idea that a well-designed simulation, much like a virtual machine, could be undetectable if constructed with formal proofs that prevent information from entering or leaving. This sets the stage for a broader discussion on the nature of hacking and the ethical considerations that come with it, as Hotz reflects on his own evolution from hardware manipulation to software exploitation.
Hotz then shifts the conversation to autonomous vehicles, emphasizing the importance of effective lane keeping as the primary value proposition for such technology. He shares his experiences working with Elon Musk on autopilot systems, revealing the complexities and challenges involved in developing stable, reliable systems. Despite initial skepticism, Hotz claims to have matched the quality of Mobileye's lane-keeping system in a short period, underscoring his belief in the potential of autonomous driving technology to revolutionize transportation.
The discussion takes a critical turn as Hotz critiques the industry's reliance on LIDAR for autonomous vehicles, arguing that it is more of a crutch than a solution for perception challenges. He highlights Tesla's data-driven approach as a competitive advantage, suggesting that real-user data collection is key to achieving level five autonomy. This perspective challenges conventional wisdom and positions Tesla as a leader in the race for fully autonomous vehicles, despite the technological lead held by companies like Waymo.
Finally, Hotz outlines Comma.ai's strategic vision, which includes a pivot towards the insurance industry. By leveraging the data collected from their autonomous systems, Comma.ai aims to offer lower premiums for safe drivers, potentially disrupting traditional insurance models. This business model aligns with Hotz's broader goal of using technology to enhance safety and efficiency, while also ensuring profitability and long-term sustainability for Comma.ai. The episode concludes with an open-ended discussion on the future of AI and its role in shaping human experiences.
Surprising moments
In-depth
Simulation and Hacking
- Hotz suggests simulations could be unfalsifiable if designed with formal proofs.
- He learned Coq to potentially create correct-by-definition simulations.
- His first iPhone hack involved physical hardware manipulation.
Autonomous Vehicles and Lane Keeping
- Effective lane keeping is seen as a primary value add for autonomous vehicles.
- OpenPilot supports 45 car makes, with the 2020 Corolla as the best fit.
- Tesla's real-user data collection offers a competitive edge.
Business Models and Insurance
- Comma.ai aims to become a car insurance company leveraging its data.
- This model could offer lower premiums for safe drivers.
- The company plans to be profitable by mid-next year.
LIDAR and Perception
- Hotz criticizes LIDAR as a crutch primarily for localization.
- Waymo is seen as having a technological lead over competitors.
- The static driving problem is considered solvable, but dynamic challenges remain.
Notable Quotes
If the simulation is designed in such a way that they did like a formal proof to show that no information can get in and out, and if their hardware is designed for anything in the simulation to always keep the hardware in spec, it may be impossible to prove whether we're in a simulation or not.
Still open
- Hotz questioned whether LIDAR's role in autonomous vehicles is justified, given its limitations in perception.