Comment by wgj

> Pretraining a GPT 4-class model from raw data requires trillions of tokens and millions of dollars in compute,

And millions of documents authored by people that weren't compensated.

The difference is consolidating all of that value into a single company.