@JulianOliver @narthur It seems to me that it would be fairly easy for them to consider that the content is junk and to treat accordingly everything coming from the same IPs. They probably keep a log of where the content used to train the LLMs is coming from (maybe with some kind of hash / pseudonymous maneer), and likely have ways to reject content from the same server IF they detect a problem from several domains linked to this server. IMO a bunch of reverse proxies / various IPs could help : they might be dumb and make it easy to polute their dataset, but probably aren't.