Blog

Why placement and routing, not arithmetic or energy, is the binding bottleneck in photonic computing, and the case for a physics-informed generative approach to chip layout.

We examine why deploying vision-language-action models on real robots remains an unsolved systems problem, and explore the inference, memory, and observability gaps between impressive demos and production deployment.

We start from Claude Shannon's foundational work and build to some ideas in modern machine learning, and explore entropy, mutual information, and their unexpected connections.