We've been telling ourselves the wrong story about large language models.
When we talk about AI systems like GPT-4, Claude, or Llama, we habitually use metaphors that ascribe agency: they "think," "decide," "reason," or even "hallucinate." These anth...
