The two are inseperable. The scale of large language models means they can only be trained by people willing to spend billions on data harvesting and compute.