>"To understand why neural language models are dangerous (and fascinating), we need to approach them as models of culture."
"I understand why researchers in a field named “artificial intelligence” would associate meaning with mental activity and see writing as a dubious proxy for it. But historical disciplines rarely have access to minds, or even living subjects. We work mostly with texts and other traces. For this reason, I’m not troubled by the part of “Stochastic Parrots” that warns about “the human tendency to attribute meaning to text” even when the text “is not grounded in communicative intent” (618, 616)."