The logical destination of LLMs

Graze, the company that do really good stuff around AT protocol user controlled algorithms, have announced that they’re trying to “Billionaire proof inference” by creating an AI cooperative called co/core.

It’s quite an interesting idea, where users share their own personal spare compute, from devices like an old MacBook, with a network of other users doing the same thing. The idea is that LLM tasks are run on each other’s machines and everyone appears to mutually benefit.

The logical endpoint to this technology is that it’ll be these small, niche networks and local specialised models, used by technology enthusiasts. Add it to the pile of other “game changers” that are now predominantly used by niche groups. The economics don’t add up for the rest and that’s even while paying users/companies are still being hugely subsidised. When the real costs hit, the ROI just simply isn’t there.

I’m genuinely glad to see stuff like this co/core initiative happening — it’s actually quite refreshing. I do worry about security, but it’s an experiment, just like the AT protocol was, so I expect that stuff to be earnestly ironed out by decent people.

Do I personally have any interest in it though? Nah, because the stink of the tech oligarchy, environmental harms, psychological harms and being told every fuckin’ day that LLMs are the “future” will very likely never pass for me, so I can’t ever seeing me being interested in using this technology in any meaningful way.

I do appreciate that people genuinely enjoy using this technology, though, so at least there is some more accessible movement away from the sweaty tech oligarchy and their infinite losses machine.