
I sit down with Karol Hausman and Kevin Black of Physical Intelligence (Pi) to unpack how they are building a general-purpose robot brain so any robot can perform any task, anywhere. We talk about the convergence of AI, robotics, and automation and what it will take to give machines true physical intelligence.
If you're interested in the future of general-purpose robots, physical AI, or the future of labor, then give this episode a listen/watch.
#robots #ai #decentralization
Chapters:
00:00 Introduction
01:06 Vision-Language-Action Model (VLA) vs Vision-Language Model (VLM)
04:42 "Taylor Swift" Pivotal Moment In Robotics
07:40 Training Robots With Natural Language Prompts
09:50 Pi's "Action Expert" Architecture
13:02 Pi’s Open Source Strategy
16:11 Perspectives On Building Hardware
22:12 Creating An Ecosystem For Physical Intelligence
24:51 Pi 0.5 Model And Open World Generalization
32:18 Hitting Diminishing Returns On Task-Specific Data
35:16 Tackling Real-Time Inference Speed in Robotics
38:33 Improving Context Length
44:53 Importance of In-House Data Collection
47:41 Opportunities For Service Providers In Robotics
48:28 The Role of Hardware in Robotics
50:00 Dealing with Edge Cases And Data Diversity
52:22 Founding Story Of Pi
55:28 Kevin’s Journey To Pi
58:49 Opportunities For Non-Technical Founders In Robotics
01:01:03 Exploring Areas For Early Deployment
01:01:44 Rapid Fire Questions on Robotics
01:01:56 One Assumption AI Researchers Get Wrong
01:06:49 Closing
Follow Jordan on X: https://x.com/jrwolfe
Links to Karol & Kevin’s Work:
Key Influences and Resources Mentioned:
Sergey Levine’s Substack - https://sergeylevine.substack.com/
Subscribe & Follow:
Subscribe to YouTube for more deep dives on decentralization, robots, and the real economy.