What AI Changes in Scientific Workflows
I have been thinking about world models in a more developmental way: not as systems that simply produce plausible outputs, but as systems that gradually build grounded understanding. My intuition is that if AI is going to matter deeply for science, it may need to learn in layers, starting from perceptual structure and moving toward causal, mathematical, and eventually scientific understanding. What interests me most is the idea that intelligence may require not just scale, but a developmental path.
Published March 30, 2026