Phrases.pdf – how well do LLM predictions compare with actual corpus data | Not Hacker News!