Can we just push LLMs aside for a minute and look at the whole AI models from outside the box? The feeling I'm getting is that the obsession with LLMs has outpaced its usefulness.