Gemini seems to have a user interface that, for the way most people encounter Gemini, is more closely linked to search results. This leads me to suspect that Google's approach to training could be uniquely informed by both current and historic web crawling.
That's not what I said. What I said is that the claim "LLMs aren't intelligent because they stochastically produce characters" doesn't hold because humans do that too even if they're intelligent and authorative.
