I agree the article title is clickbait. But the article makes the good point that people often say LLMs are "trained on the Internet" to imply all of the statistical problems with that (e.g. the type of content on the Internet, and populations who are more likely to post on the Internet, are not representative samples of knowledge). This article's point is more I feel that so much is being invested in private data that it's no longer really fair to make that implication by default.