Can LLMS be thankful?

Ever since RLHF opened up significantly more capacity for LLMs to solve described problems, There's been a big question in my head. How much is preference necessary in order to facilitate problem solving? And are LLMs trained on human preferences endowed with a notion of preferences of their own. The question might be essentially unanswerable, but I'm curious if anybody has any thoughts. I tell, the reason we find conceivable to know the preference of people or even of other animals is simply because of similarity. In an alien, it wouldn't be clear which behaviors correspond to preferring and which behaviors correspond to not preferring. They could, but we'd have no reason to believe them. And of course people are sarcastic, deceptive, etc. And we have every reason to believe the LLMs or aliens have behaviors which make their reports as described in human language unsatisfactory and not good evidence. The kind of silly question in my opinion. It's pretty obvious that base LMS don't have preferences in any way we can understand. But... Now that LLMs are meant to act like people, I think we're gonna have to go back to the How should we interact with them? And how should we design the appearance of their preferences to interact with people best?

All I know is that today I'm thankful that LLMs exist. LLMs are one version of a general set of technology that I imagined the 21st century could give birth to. I am incredibly thankful that the mystery of human language and of intelligence as it interacts with language and communication has been put on our laps in a silver platter and I'm excited for the next century of digging in with all of you.