1001ud

On over-applying logic to relationships, and the regret that followed.

The tradeoffs of exposure, institutional gatekeeping, and compounding audiences.

A verified finite-difference acoustic reference compared against a WavePINN-NIF surrogate, with held-out evaluation showing where the neural field fails.

A look into separating the roles of machine learning models for visual content, focusing on using diffusion for surface appearance while relying on predictive and physics-informed models for stable geometry and motion.

An overview of the neural-surrogate projects in this section and the rest of the repo: cloth, waves, manifolds, motion, electromagnetics, trajectories, and real-time surrogate graphics.

Traditional simulation is expensive. A consideration of the compute requirements of solvers and the compounding gains of physics-informed neural surrogates.

A reflection on trained neural networks as compressed, parameterized functions rather than explicit data.

Why the most interesting question in language technology is what happens when you point it at people the mainstream system wasn't designed for.

On Yung Lean, building from outside the culture, and why the outsider position is not a disadvantage to overcome.