> but we really want it to be internal and automatic. Why? We have specialized semi-isolated lobes of our brain. Some "external to the LLM weights" guiding sys

> but we really want it to be internal and automatic. Why? We have specialized semi-isolated lobes of our brain. Some "external to the LLM weights" guiding system just needs to be transparent to us. It doesn't need to be internal to the LLM/weights. > because what controls the internal thought process is itself not necessarily language-based Most of the interesting AI things can't write a sentence (weather simulation, protein folding, inverse kinematics, driving cars, etc...). Language isn't a requirement, only some sort of meaningful mapping to the latent space of each is. I think one could claim that "meaningful latent space mapping" is the language of neural nets, including those in the brain.