AI Models Lack Physical Understanding for Hardware Engineering Tasks
We are heading into a memory-constrained world. Every wafer that goes to AI memory is a wafer that doesn't go to robotics, phones, or PCs. And relief isn't coming until 2027 at the earliest. I don't think we're getting 20 million robots in five years, and the demo reels won't tell you why. Memory, actuators, and safety all need a few more cycles. 5/
Caitlin Kalinowski@kalinowski007Today's frontier models do not understand friction, weight, surface texture, or contact pressure. To do real CAD, electrical engineering, or robotics design, you need world models that do. Whoever builds "Codex for hardware engineering" will own a chunk of the next decade. It does not exist yet. 4/
@lennysan We also got into: • the actuator supply chain • why I left OpenAI • why I'm skeptical of humanoids • what Apple's "back of the cabinet" obsession still teaches every hardware team • how I hire AI-native engineers
Full episode 6/
Caitlin Kalinowski@kalinowski007We are heading into a memory-constrained world. Every wafer that goes to AI memory is a wafer that doesn't go to robotics, phones, or PCs. And relief isn't coming until 2027 at the earliest. I don't think we're getting 20 million robots in five years, and the demo reels won't tell you why. Memory, actuators, and safety all need a few more cycles. 5/
@lennysan https://www.youtube.com/watch?v=G5WTgB87rYQ
Caitlin Kalinowski@kalinowski007@lennysan We also got into: • the actuator supply chain • why I left OpenAI • why I'm skeptical of humanoids • what Apple's "back of the cabinet" obsession still teaches every hardware team • how I hire AI-native engineers Full episode 6/