> Frontier science looks different today. It's rooted in model weights and GPUs. It is flooded with token spend and agentic loops. It blooms in data centers.

This seems like handwaving to me. Even the actual frontier science that is using ML (e.g. AlphaFold) isn't based on "token spend" or "agentic loops". I personally cannot think of a single example of frontier science that is rooted in LLMs. I am sure there are a few examples, but the idea that frontier science has somehow completely shifted its trajectory and methods based on LLMs feels like quite a stretch to me.

Got any counterexamples to show me I'm wrong?