A posttraining data commons and the infrastructure surrounding it is the next great open source project IMO, akin to Postgres or Linux.

The Marin project from Stanford (and I believe now ran by Openathena, the authors of this blog post) is really promising and cool: https://marin.community/