Language models cannot reliably distinguish belief from knowledge and fact | Not Hacker News!