Yes, definitely. We and other groups are building differentiable physics based models based on our current ones.

Show Post

We still need to understand what the models are doing both mechanistically and conceptually because the space of possible options is so big. With the right compute and pipelines we can be more systematic about testing what works and what doesn’t.

Show Post

Drew Rasmussen in three postseason starts: 21 innings 3 runs (all earned) 1.29 ERA 23 strikeouts 3 walks 13 hits Absolutely dominant stuff.

Show Post

Ultimately the observational data we have doesn’t capture all the physical phenomena responsible for weather, and the physics we have doesn’t explain everything we see in the data. Frame the broad structure with physical understanding and give ML the flexibility to fill in details.

Show Post

As we have seen with large language models, multiple forms of training/refinement and feedback loops are needed to find new frontiers in performance. Building from but also critiquing our current conceptual models of our physical models will identify many areas for improvement.

Show Post