K (Physical Intelligence)

K is the Physical Intelligence researcher interviewed in 146. 对Physical Intelligence柯丽一鸣4小时访谈：Pi的开源模型研究，机器人的江湖、族谱与主角. The source identifies him as a reinforcement-learning-focused researcher and uses his path through programming, psychology, economics, game theory, machine learning, imitation learning, and robotics to explain why current robot intelligence is a system problem rather than a single-algorithm problem.

K’s technical position is pragmatic. He values real-machine performance, learned methods, and Robot Experience Data, but also keeps the traditional robotics insistence on physical success and deployment constraints. In the episode, this makes him a bridge between old robotics control/planning culture and the current Vision Language Action Models and Robot Reinforcement Learning wave.

The episode also gives K a personal intellectual profile: he writes science fiction, thinks about games and future society, uses AI agents such as Claude Code to change his work process, and is skeptical of simple answers to AI doom, consciousness, robot race, and life-meaning questions.

Connections

• Physical Intelligence and Physical Intelligence Pi Model Series — company and model program K explains.
• Robot Reinforcement Learning, Robot Experience Data, and Robot Generalization Performance Tradeoff — core technical ideas in his account.
• Robot Form-Factor Pragmatism — his practical stance on humanoid versus non-humanoid robots.
• Human Judgment Under AI and AI Organization Design — adjacent reflections on AI-agent work, expert trust, and responsibility.