What is language for?
Jennifer Michalowski,
McGovern Institute,
Jun 28, 2024
This is a bit funny but pretty important. It's funny because the central claim is being used as an argument against the use of large language models (LLM) for artificial general intelligence. Here's the claim: "Drawing on evidence from neurobiology, cognitive science, and corpus linguistics, researchers make the case that language is a tool for communication, not for thought." I think that's quite right. But while LLM are limited in this way, neural network based AI in general is not. And even more importantly, the claim refutes a theory known as the Physical Symbol System hypothesis, which implies that that human thinking is a kind of symbol manipulation, and which forms the basis for a lot of the cogitive theory out there today (including, ironically, most critics of LLM). This article is a summary of the paper printed in Nature (paywalled, but there's a copy posted here). See also this Hacker News discussion. Via Benjamin Riley in a post that I think contains a number of errors, who is referenced in this discussion on the Learning Engineering discussion list.
Today: 4 Total: 966 [Direct link] [Share]
