The Princeton mathematician was arguing about the limits of large language models. He says human thought starts as intuition and only later becomes words, so a model trained only on words lacks that starting point.
…on what we know so far. And also, you can go to more specific things. Like, for example, it's pretty clear that our thoughts, I'm not talking about all thoughts, but the ones that matter. Predate the language, right? We have an intuition about something and then we find the words to explain it.
It's like a proto-language.
I mean, that's something that AI doesn't have. I mean, AI is completely based on guessing the next word, the next token. In other words, it's completely based on words.
Yeah.
Language. Yeah. On language. But as I said, I mean, I think our thoughts predate the language. We have intuitions about things, which
we have physical embodied emotion. Exactly. Einstein falling in space. I don't have to tell, I can't describe that to you, but you know exactly what it feels like to go in a roller coaster or fall in love.