Elon Musk, Tesla CEO, answers audience questions with Tesla AI and engineering team members at a Tesla AI event about Dojo, Autopilot, simulation, and the Tesla Bot.
Musk said Tesla might license its AI system to other carmakers, while Dojo engineer Bill described scaling challenges around memory, bandwidth, and keeping computation local. Musk said Dojo’s initial job would be training on large volumes of driving video, with success measured by whether Tesla’s software team preferred it to its GPU cluster. Ian Glow said Tesla planned to extend its driving simulator to support Optimus and potentially become a broader simulation platform.
Musk described Tesla’s driving system as primarily vision-based, with audio used for cases such as emergency vehicles and spoken commands. He said Tesla trains on data from many countries but focuses development on the United States before adapting more broadly. The team described combining neural networks with explicit physics and planning methods, and using targeted simulated clips to cover rare situations while evaluating performance on real-world video.
Musk said the Tesla Bot was not a response to manufacturing needs, though manufacturing expertise would matter in keeping it affordable. He framed its first uses as repetitive, dangerous, or tedious work, and said the initial design would have five fingers for handling tools and bags. On factory automation, he cautioned that machine learning is difficult and should be used only when a task calls for it.
A Tesla AI team member said larger models can improve performance when enough training data is available, especially in auto-labeling, though onboard networks face latency limits. Musk described using large models to create labels and then distilling their output into networks that run in cars. He also said Tesla accounts for multiple possible actions by other road users, slows down when uncertain, and uses shadow-mode comparisons with human drivers to assess new behavior.
Musk said Tesla expected its driving networks to improve with more data and newer architectures, and claimed Hardware 3 could reach safety well beyond human driving. He said a future Full Self-Driving Computer 2 might offer about four times the capability, and that Hardware 4 would include a next-generation camera. On AI risk, he characterized Tesla’s current work as narrow applications and said broader superhuman intelligence deserved concern.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
RoboStrategy head of research Scott Walter, an aerospace engineer and robotics-company co-founder, joins host Herbert to discuss Tesla Optimus V3’s design, production readiness, and remaining engineering challenges.
Elon Musk, CEO of Tesla, discusses Tesla’s China operations, AI, humanoid robots, and space exploration with a CGTN host at Tesla’s Global Engineering Headquarters.
Andrew King, CEO of RoboStrategy, discusses humanoid robotics and investment with John Koetsier on NEXT.
Scott Walter joins The Humanoid Hub host to analyze Tesla’s Optimus hand and production plans, Figure’s robot output, and humanoid robots’ performance in the Beijing half marathon.
Host speaks with Scott Walter, a roboticist known for humanoid hardware analysis, on The Humanoid Hub about Tesla Optimus hand patents and what they reveal about the hand’s design.
Matt Gilbert, AI and data director at Synapse, discusses the state and future of physical AI with Jonas on Tech First.
