In this episode, Prakash Narayanan and Nathan Labenz move from frontier-model safety questions to the practical limits of robots in the real world. Timothy Lee joins to discuss humanoid robot reliability, Tesla versus Waymo, robotaxis, and the role of vision-language-action models in getting machines to work outside the lab.
Dr. Jean Nehme then explains soft robotics, intelligent materials, and how biology can inspire robotic systems that sense, adapt, and change shape. The conversation also touches on AI safety, secure coding, rogue agents, and the broader question of how physical AI may evolve beyond the humanoid form.
Show Notes
Prakash Narayanan and Nathan Labenz are joined by Timothy Lee and Dr. Jean Nehme for a wide-ranging conversation about GPT-6 safety restrictions, hidden reasoning, rogue agents, robot reliability, and the bottlenecks in AI-assisted coding and robotics. The episode then turns to soft robotics, biology-inspired materials, and why future physical AI may move beyond the humanoid form before closing on AI regulation, finance, and tail risk.
Chapters
(0:00) Can AI hide its reasoning?
(0:25) Humanoids can fall in homes.
(1:17) Robot hands are the bottleneck.
(1:52) AI takeover needs no robots.
(2:23) The AGI era begins
(5:00) Frontier Math and superintelligence
(7:05) AI-built 3D worlds
(8:32) Code benchmark gap
(9:42) GPT-6 safety restrictions
(14:11) Hidden reasoning and monitoring
(16:49) AI game design demos
(19:55) Biotech video generation
(23:50) Rogue agents go online
(30:25) Frontier defense factory
(33:02) Secure AI-assisted coding
(34:05) Timothy Lee and robotics
(41:06) Industrial robot design
(43:15) Tesla versus Waymo
(47:10) Tesla FSD experience
(56:39) Vision-language-action models
(1:03:22) Robot reliability
(1:14:27) AI safety and control
(1:15:00) Rogue AI agents
(1:18:58) Humanoid robot control
(1:20:55) Meet Dr. Jean Nehme
(1:25:18) Biology as a robotics blueprint
(1:33:08) Cells become robotic bodies
(1:36:47) Computation in soft materials
(1:43:05) Beyond the humanoid form
(1:46:39) Manufacturing intelligent membranes
(1:52:08) Physical AI beyond metal
(2:06:19) The robot skills paradox
(2:10:22) Closing on AI’s turning point
(2:11:27) Health AI’s lifesaving upside
(2:13:51) AI pause and dangerous swarms
(2:16:42) The point of no return
(2:21:26) Why an AI pause is hard
(2:24:18) Pause without economic collapse
(2:34:30) GPT-3 access and AI winners
(2:36:48) Automatic software formalization
(2:45:18) AI takeover through finance
(2:47:48) Data centers versus housing
(2:49:32) Capitalism and AI tail risk
(2:52:52) Paperclip maximizers and finance
(2:56:23) Escaping paperclip maximization
(2:57:44) Using Astra and Fable 5.1
(2:59:05) Why AI forgetting matters
Guests:
Dr. Jean Nehme — Founder & CEO, morph (LinkedIn)
Timothy Lee — Founder, Understanding AI (𝕏 | LinkedIn)
