Language Log » Stochastic parrots extended
Mark Liberman,
Language Log,
Jun 27, 2024
Mark Liberman links to and discusses briefly a paper from Philip Resnick called Large Language Models are Biased Because They Are Large Language Models which suggests that large language models (LLM) are unable to distinguish between statements about what a word a word means and statements about the thing a word refers to (that is, between facts about meanings and facts about the world). Because of this, Resnick argues, LLMs are inherently biased in a way that can't be fixed. My question is, is it true that "LLMs... have no way to distinguish among distributional patterns that arise from definitions or meaning" versus statistical generalizations. That's not so clear to me. Image: Florian Pestoni.
Today: 3 Total: 351 [Direct link] [Share]
