Comment by mydpy

Right, but this is still impressive given how quickly Databricks created and open-sourced the dataset. I would expect more improvements in the future.