People are supposed to communicate symbolically with LLMs too

divide and conquer :)

I think a more specific denominator you might wanna look into is free energy minimization as you mention in your first example. I really liked reading Active Inference and What Is Life? on the subject.

I'm working on something tangentially related [1] but by sourcing my Google search history data. It's surprising how LLaMA 3.1 8B is pulling most of the weight in my case too.

[1] https://github.com/enclaveid/enclaveid

Unless you have a massively parallel use case and a small, specialized model (which seems the general direction of the industry anyway)

Off-topic: I really reasonate with with an old comment of yours, ID 32330140. Commenting here as I couldn't find any another way of contacting you, could you send me an email? Thanks