I wonder if there is way local small LLMs can complement each other in away that the sum-total yields a much more performant LLM

Perhaps some radical MoE where you download _exactly_ the components you need as you need them. Currently MoE is switched usually on per-token per-layer basis, so you need all weights locally. But e.g. Apple made one which pre-selects all experts based on prompt embedding. That might be further scaled up - e.g. predict exactly what's needed

I don't understand why no labs create dedicated models per industry/expert. E.g. physics, electronics, chemistry, etc. Each model would be much smaller and better suitable for running locally. Everyone is trying to cram everything into a single model.

Perhaps something similar to speculative decoding.

Speculating Experts Accelerates Inference for Mixture-of-Experts: https://arxiv.org/abs/2603.19289

Sort of like how ants in a colony produce a working "society" that no individual ant could muster.

Can someone explain to me what is their "prompting-scaffolding" to make it work ?

"This is a general-purpose LLM. It wasn’t targeted at this problem or even at mathematics. Also, it’s not a scaffold. We have not pushed this model to the limit on open problems. Our focus is to get it out quickly so that everyone can use it for themselves." - Noam Brown (OpenAI reasoning researcher) on X