https://www.nature.com/articles/s41559-022-01784-1

Big LLMs are far more complex, but they’re also made be people who don’t have a clue what they’re doing, so in the future we’ll probably have much smaller ones.

It’s right there in the article

https://en.wikipedia.org/wiki/Gibbard's_theorem

No, https://en.wikipedia.org/wiki/Arrow's_impossibility_theorem

You’re wrong about everything

How much extra storage and bandwidth would it use? It’s just text. It’s clear that you don’t like the idea, but I don’t find your reasons convincing.