It makes a lot of sense, but with the way organizations such as Internet Archive are saving webpages from Reddit, wouldn't it be feasible to train your models off of those sites to circumvent any API charges?