Jesse Levinson on AI architecture
From Society Signature Symposium 2025 - Jesse Levinson and Maya Ajmera · · SocietyforScience
“We are using a much more hybrid architecture that does work. We have neural nets all over our system, but that's not the only thing we have to keep people safe. We are directly measuring the 3D structure of everything around us and explicitly encoding in our software to do everything you can to not run into things.”
On , Jesse Levinson, Cofounder at Zoox, spoke about AI architecture during Society Signature Symposium 2025 - Jesse Levinson and Maya Ajmera on SocietyforScience.
About Jesse Levinson
Jesse Levinson, co-founder and CTO of Zoox, stated in mid-2025 that the company was ready to deploy its purpose-built robotaxi on public roads in San Francisco and on the Las Vegas Strip, describing the vehicle as having no manual controls and being designed for riders. He noted that Zoox had achieved an internal safety readiness bar requiring the vehicle to be "many, many times better than humans" at avoiding collisions. Levinson attributed the decade-long development timeline to the difficulty of building a comprehensive safety case, which he described as a set of quantitative arguments proving the vehicle is safer than a human driver)Skip to content Levinson has contrasted Zoox’s approach with that of Tesla, stating that Tesla lacks the technology for driverless operation without a human and that its system is "about 100 times less safe than a human." He emphasized that Zoox uses a hybrid architecture combining neural networks with direct 3D structure measurement and explicit safety encoding, rather than relying solely on end-to-end machine learning. Levinson also discussed Zoox’s hiring philosophy, saying the company values critical thinking over domain-specific knowledge, and noted that most new hires have not previously worked on autonomous vehicles.