This article highlights a really interesting paper (19 page PDF) that "showed that language models, the engines of commercial AI chatbots, show strong signal correlations with the human language network, the region of the brain responsible for processing language." Even more interesting, this correlation doesn't last - as the language models become more capable, the correlation gradually fades, suggesting that (and these are my words) the models are optimizing beyond langauge. To me, this offers the tantalizing possibility that language, rather than being the thing that uniquely enables cognition, is currently the thing that is limiting the capability of (human) cognition. Anyhow, you don't have to subscribe to my interpretation to find this article and paper worthwhile. Found via the TWIT weekly podcast (about an hour in).
Today: Total: [Direct link] [Share]
