To model human linguistic prediction, make LLMs less superhuman

Open in new window