I could get the same thing out of "curl https://en.wikipedia.org/wiki/Corn" but curl doesn't understand what corn is any more than Claude does. Claude doesn't understand corn any more than Wikipedia either. Just like with Wikipedia, everything Claude outputs about corn came from the knowledge of humans which was fed into it by other humans, then requested by other other humans. It's human understanding behind all of it. Claude is just a different way to combine and output examples of human thoughts and human gathered data.

You know it when you see it, but it seems to like an objective definition that stands up to adversarial scrutiny. Where are the boundaries between knowing and repeating? It can be a useful idea to talk about, but if I ever find myself debating whether "knowledge" or "understanding" is happening, there will probably not be any fruitful result. It's only useful if everyone agrees already.

I guess that's basically the idea of the Chinese Room thought experiment.