[tasty_freeze]

People often say that LLMs aren't really thinking because they are just producing a stream of words (tokens really) reflexively based on some windows of previous text either read or from its own response. That is true.

But I have the experience when talking of not knowing what I'm going to say until I hear what I've said. Sometimes I do have deliberative thought and planning, trialing phrases in my head before uttering them, but apparently I'm mostly an LLM that is just generating a stream of tokens.

[Workaccount2]

This is something that is easily observable by anyone at virtually any moment, yet at the same time is something that escapes 99% of the population.

When you are talking to someone in normal conversation, you are both taking in the words you are saying at the same time.