If you inspect the Chain of Thought summaries, the LLM often knows full well what it is doing.
That's not knowing. That's just parotting in smaller chunks.
alt Hacker News
That's not knowing. That's just parotting in smaller chunks.