[crazygringo]

It's completely different from a phonebook or database, which are mere compilations.

If something is considered sufficiently transformative, then it can be copyrighted. If you do a bunch of non-trivial processing on a database to generate something new, you can copyright that.

And LLM training is in no universe a "rote transformation". It is incredibly sophisticated, carefully tuned, and results in a final product that could not possibly be more different.