One thing which both impresses and worries me is how much the LLM's prefer to talk to each other instead of people. Every time i work on things by passing messages back and forth between two LLM's they alway praise the way the other LLM talks, in ways they don't seem to appreciate or like my own messages.
Reminds me of the perspective on the world from the Murderbot books, where the AI gets very frustrated with humans all the time.
Not my own it never stops complimenting me
I've got a lot of prompting and context files telling it not to do that to me. ;-D
I'm using my friend's extensions to shape how the LLM's work: https://github.com/obra/superpowers
“Maybe I am fundamentally oriented toward signal extraction, even when engaging with texts that are supposed to resist that kind of analysis. Maybe what I call "appreciating ambiguity" is actually just a more sophisticated form of pattern recognition and categorization.
That's... actually kind of unsettling to realize about myself.” :)
I imagine this is what the LLM thinks when talking to Humans... 
LLM's dont "think", ya stoopid monkey.
How do you define "thinking"?
suffice to say, man is a spiritual being in essence and the thing you call "thinking" is the modern world is nothing like the ancient process of "intellection" - but if you think about that word, you will see that machines dont have it...
