Imagine you have someone polluting your training data every day. That's what happens when you scrape any tech forum today.

The short version is that llm trainign data is the lowest quality data you are likely to see unless you engage in massive potential copyright infringement.