>>280347892Idk about you but I can go from concept to speech while understanding both. Humans have something called understanding, we use speech to translate a concept into a form that can be communicated.
You're not just trying to predict what the next word in a conversation should be through some sort of probabilistic model, you're trying to communicate some sort of idea.
LLMs don't do this, they're a purely probabilistic model with no actual thinking or understanding.
https://arxiv.org/abs/2506.21521
This paper is a great example. Unless you're talking to someone with very advanced Alzheimer's you're not going to get these kinds of answers. If someone understands a concept then they'll also be able to solve a simple problem by applying it.