Same harness trains optical encoders; paradigm generalizes to any controllable physical system

GordonWetzstein · x · 2026-11-20

The same harness also trains an optical image encoder, and more capable agents yield better results. Since the agent only needs a way to set device parameters and observe outputs, the approach extends to any physical system that can be controlled and measured.

Related event: Stanford Trains Optical Neural Networks In Situ with Coding Agents(2 posts)→

Original post →

More from Research

Research channel →