People paying reddit aren’t training their models by scraping the web ui.

The original comments are still in the database, which is what you’d pipe into an LLM.