Episode
#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
3:13:27
Published: Thu Oct 22 2020
Description
George Hotz (geohot) is a programmer, hacker, and the founder of Comma.ai. Please support this podcast by checking out our sponsors: - Four Sigmatic: https://foursigmatic.com/lex and use code LexPod to get up to 40% & free shipping - Decoding Digital: https://appdirect.com/decoding-digital - ExpressVPN: https://expressvpn.com/lexpod and use code LexPod to get 3 months free EPISODE LINKS: Comma.ai's Twitter: https://twitter.com/comma_ai Comma.ai's Website: https://comma.ai/ George's Instagram: https://www.instagram.com/georgehotz George's Twitch: https://www.twitch.tv/georgehotz George's Twitter: https://twitter.com/realgeorgehotz Comma.ai YouTube (unofficial): https://www.youtube.com/channel/UCwgKmJM4ZJQRJ-U5NjvR2dg PODCAST INFO: Podcast website: https://lexfridman.com/podcast Apple Podcasts: https://apple.co/2lwqZIr Spotify: https://spoti.fi/2nEwCF8 RSS: https://lexfridman.com/feed/podcast/ YouTube Full Episodes: https://youtube.com/lexfridman YouTube Clips: https://youtube.com/lexclips SUPPORT & CONNECT: - Check out the sponsors above, it's the best way to support this podcast - Support on Patreon: https://www.patreon.com/lexfridman - Twitter: https://twitter.com/lexfridman - Instagram: https://www.instagram.com/lexfridman - LinkedIn: https://www.linkedin.com/in/lexfridman - Facebook: https://www.facebook.com/LexFridmanPage - Medium: https://medium.com/@lexfridman OUTLINE: Here's the timestamps for the episode. On some podcast players you should be able to click the timestamp to jump to that time. 00:00 - Introduction 07:02 - Will human civilization destroy itself? 09:49 - Where are the aliens? 14:36 - Tic Tac UFO and Bob Lazar 17:04 - Conspiracy theories 19:07 - The programming language of life 23:28 - The games that humans play 31:58 - Memory leaks in the simulation 34:29 - Theories of everything 36:14 - Ethereum startup story 44:02 - Cryptocurrency 53:28 - Self-help advice 57:08 - Comma.ai 59:02 - Comma two 1:07:50 - Tesla vs Comma.ai 1:16:53 - Driver monitoring 1:30:34 - Communicating uncertainty 1:32:22 - Tesla Dojo 1:38:50 - Tesla Autopilot big rewrite 1:45:09 - How to install the Comma Two 1:49:44 - Openpilot is Android & Autopilot is iOS 1:58:59 - Waymo 2:10:12 - Autonomous driving and society 2:12:24 - Moving 2:15:29 - Advice to Startups 2:28:32 - Programming setup 2:31:32 - Ideas that changed my life 2:39:37 - GPT-3 2:42:57 - AGI 2:47:00 - Programming languages that everyone should learn 2:53:33 - How to learn anything 2:56:05 - Book recommendations 3:04:28 - Love 3:06:17 - Psychedelics 3:08:38 - Crazy
Chapters
Kama AI founder explains how data collected by companies like comma AI and Tesla autopilot can help design safe semi-autonomous vehicles for real-world conditions, while his work at MIT led him to study the human side of autonomy which is much richer than what can be studied in the lab.
00:00 - 02:44 (02:44)
Summary
Kama AI founder explains how data collected by companies like comma AI and Tesla autopilot can help design safe semi-autonomous vehicles for real-world conditions, while his work at MIT led him to study the human side of autonomy which is much richer than what can be studied in the lab.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
In this podcast, Lex and his guest Adi Shamir discuss the history and future of cryptography, the value of privacy, and the role of technology companies in digital security.
02:44 - 09:01 (06:17)
Summary
In this podcast, Lex and his guest Adi Shamir discuss the history and future of cryptography, the value of privacy, and the role of technology companies in digital security.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The human mind and our technology can only grasp a certain slice of the universe we live in.
09:01 - 14:35 (05:33)
Summary
The human mind and our technology can only grasp a certain slice of the universe we live in. There may be phenomena we cannot even comprehend yet, hence why we only have assumptions about the mysteries of the universe.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The prevalence of deepfakes and fake news has led to a dilution of truth and a rise in conspiracy theories.
14:35 - 20:47 (06:11)
Summary
The prevalence of deepfakes and fake news has led to a dilution of truth and a rise in conspiracy theories. This poses a danger as it can be used to hide information and manipulate public perception, as seen with the case of the Navy pilot David Fravor.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The technical and ethical costs of experimentation on a molecular level for the creation of vaccines against COVID and other illnesses raises concerns surrounding animal and human involvement.
20:47 - 24:29 (03:41)
Summary
The technical and ethical costs of experimentation on a molecular level for the creation of vaccines against COVID and other illnesses raises concerns surrounding animal and human involvement. The discussion also touches on the possibility of hacking the simulation.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The obsession with efficiency in video games has led players to sacrifice enjoyment and exploration to maximize their experience points.
24:29 - 28:18 (03:49)
Summary
The obsession with efficiency in video games has led players to sacrifice enjoyment and exploration to maximize their experience points. A sub game within the bigger game has emerged where players slaughter cows repeatedly to gain the most experience points.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The idea of living a long time might seem appealing, but it could also make life feel more meaningless as you start to realize there's not enough time to learn and experience everything.
28:18 - 31:26 (03:08)
Summary
The idea of living a long time might seem appealing, but it could also make life feel more meaningless as you start to realize there's not enough time to learn and experience everything.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The first step towards knowing how to get power from the knowledge of being in a simulation is to identify indications of such power.
31:26 - 35:59 (04:32)
Summary
The first step towards knowing how to get power from the knowledge of being in a simulation is to identify indications of such power. Testable predictions, like those in Stephen Wolfram's theory of hypergraphs, make the concept interesting from a computational perspective.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
Vitalik Buterin explains the need for a Layer Two scaling solution for Ethereum, which involves op codes using varying amounts of gas to prevent DDoS attacks, and a computational layer that runs alongside and is secured by Layer One.
35:59 - 39:23 (03:23)
Summary
Vitalik Buterin explains the need for a Layer Two scaling solution for Ethereum, which involves op codes using varying amounts of gas to prevent DDoS attacks, and a computational layer that runs alongside and is secured by Layer One.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
This episode discusses the process of securing Ethereum blockchain by sandboxing code and replacing system calls with calls to the hypervisor, which allows for safe execution of smart contracts.
39:23 - 43:57 (04:33)
Summary
This episode discusses the process of securing Ethereum blockchain by sandboxing code and replacing system calls with calls to the hypervisor, which allows for safe execution of smart contracts.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The idea of groups reaching consensus in a decentralized yet formulaic way is a powerful idea from cryptocurrencies, as disputes in traditional contracts are interpreted by lawyers, making the process messy and expensive, while smart contracts can use cryptography to check for the presence of two out of three keys.
43:57 - 47:53 (03:56)
Summary
The idea of groups reaching consensus in a decentralized yet formulaic way is a powerful idea from cryptocurrencies, as disputes in traditional contracts are interpreted by lawyers, making the process messy and expensive, while smart contracts can use cryptography to check for the presence of two out of three keys.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
In distributed systems, leaders help drive the mission and vision of a project and put a face to it.
47:53 - 54:05 (06:11)
Summary
In distributed systems, leaders help drive the mission and vision of a project and put a face to it. In the crypto world, leaders often become superstars due to their expertise and knowledge.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
Despite reaching arbitrary milestones, self-driving cars still seem far from being solved.
54:05 - 58:45 (04:39)
Summary
Despite reaching arbitrary milestones, self-driving cars still seem far from being solved. While exploring new territories is important, struggling with tasks outside of one's expertise can be difficult.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
Open Pilot is a hardware system that controls car functions such as dealing with a banked road and counter steering.
58:45 - 1:03:46 (05:01)
Summary
Open Pilot is a hardware system that controls car functions such as dealing with a banked road and counter steering. It is like cruise control but with a neural network executing all of the driver's wishes.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The use of reinforcement learning in self-driving cars allows them to train completely end to end on user data without the need for lane detection or object detection tasks.
1:03:47 - 1:07:14 (03:26)
Summary
The use of reinforcement learning in self-driving cars allows them to train completely end to end on user data without the need for lane detection or object detection tasks.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
Comma AI uses end-to-end learning on actual driving data from users to train its system for successful control of the vehicle while maintaining the lane and longitudinal control.
1:07:14 - 1:14:52 (07:37)
Summary
Comma AI uses end-to-end learning on actual driving data from users to train its system for successful control of the vehicle while maintaining the lane and longitudinal control. Their approach differs greatly from the rest of the self-driving car industry, including Tesla.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The improvement of self-driving cars could lead to people over-trusting the system, resulting in the need for a tight driver monitoring policy to ensure safety on the road.
1:14:52 - 1:21:43 (06:51)
Summary
The improvement of self-driving cars could lead to people over-trusting the system, resulting in the need for a tight driver monitoring policy to ensure safety on the road.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The debate on whether to add driver monitoring systems (DMS) in autonomous cars continues, as some argue it is the wrong solution for improving safety while others believe it is a necessary feature for ensuring driver attention and intervening when needed.
1:21:43 - 1:26:45 (05:01)
Summary
The debate on whether to add driver monitoring systems (DMS) in autonomous cars continues, as some argue it is the wrong solution for improving safety while others believe it is a necessary feature for ensuring driver attention and intervening when needed.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The use of driver monitoring systems in autonomous vehicles could provide significant benefits, as it allows for training of the human and the bar for success in detecting attention is much lower than for external perception and control.
1:26:45 - 1:30:40 (03:55)
Summary
The use of driver monitoring systems in autonomous vehicles could provide significant benefits, as it allows for training of the human and the bar for success in detecting attention is much lower than for external perception and control. However, there is still uncertainty and a need for high accuracy in these systems.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
Nvidia has been dominating the chip market in various areas, such as autonomous driving, but competitors like Tesla and Apple are developing their own chips for specific purposes to avoid putting all their stock in Nvidia.
1:30:40 - 1:37:19 (06:38)
Summary
Nvidia has been dominating the chip market in various areas, such as autonomous driving, but competitors like Tesla and Apple are developing their own chips for specific purposes to avoid putting all their stock in Nvidia. Some believe Nvidia's focus on productizing their chips may shoot themselves in the foot and cause others to challenge their stronghold.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The use of end-to-end deep learning for autonomous vehicles has potential to make predictions more holistic and annotation tasks easier, but also presents challenges in creating accurate models that are able to perform numerous tasks.
1:37:20 - 1:41:14 (03:54)
Summary
The use of end-to-end deep learning for autonomous vehicles has potential to make predictions more holistic and annotation tasks easier, but also presents challenges in creating accurate models that are able to perform numerous tasks. Additionally, the use of these models could lead to new tools and an ecosystem for autonomous vehicle development.
Episode#132 – George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets
PodcastLex Fridman Podcast
The better technology always wins and lying always loses.
1:41:14 - 2:49:43 (1:08:29)
Summary
The better technology always wins and lying always loses. The question remains, should one essentially lie to raise money like in the case of Fyre Festival, or have a minimum viable product, users, and growth numbers as traditional Silicon Valley applies?