I'm hoping they fail to incentivize using legal, open, and/or licensed data. Then, thry might have to attempt to train a Claude-class model on legal data. Then, I'll have a great, legal model to use. :)