I don't think the issue is LLMs inherently, it's the misapplication. Kagi has siloed the LLM content into the 'Quick Answer' and 'Assistant' sections, and it generates on the fly from search results. (Plus, the 'Expert' LLM cites the references it used.) I think the issue will come when there isn't a clear delineation between real and artificial text, or when artificial text is presented more prominently than real text, as in the article.