Physical Intelligence, a robotics company in San Francisco, operates in a stark, concrete space filled with both casual lunch areas and serious robotic experimentation. The company is developing general-purpose robotic models that can learn from a variety of tasks, like folding clothes or peeling vegetables. This learning process uses data collected from various locations, including homes and warehouses, where robots are tested in real-world environments. Sergey Levine, a co-founder and UC Berkeley professor, compares their approach to ChatGPT for robots, emphasizing the importance of good intelligence over hardware limitations. Lachy Groom, another co-founder and a former Stripe employee, has led the company to raise over $1 billion, valuing it at $5.6 billion. His focus is on building a strong foundational model rather than rushing to commercialization. Investors include Khosla Ventures and Sequoia Capital, but Groom admits he doesn't provide timelines for profitability. Instead, the strategy hinges on cross-embodiment learning, which allows new robot platforms to leverage existing knowledge, reducing the costs associated with onboarding new systems. The company is already collaborating with businesses in logistics and food production to test its automation capabilities, aiming for broad applicability across different tasks and platforms.