Hacker News Reader

LLMs are great and have many use cases but ultimately they’re just very talented bullshit machines.

This is due to the way they were trained: predict the most likely next word. In a difficult context the most likely next word is a filler word.