Jonathan Stray @jonathanstray.com · Sep 3 The Octopus thought experiment sucks. If your instinct is like "LLMs can't learn anything about the actual world from mere text" AND you are intellectually curious, here is your reading list. 1/4
8 likes 1 replies
?
Replies
First, can LLM output "mean" anything? Do the words actually "reference" objects in the world even if the machine has no intentionality? Yes, the externalist view of language argues, because the training text has a natural history. Do Language Models' Words Refer? arxiv.org/abs/2308.05576 2/4