Post #3977508

@woozle@toot.cat I mean, most people just don't need an LLM for anything. The people for whom LLMs actually are useful can use open weight models locally for a fraction of the electricity and those models only need to run while inference is happening. Meanwhile, the "AI" companies can only make money if people can't do that. So they're trying to push everyone to use models that need a terabyte of vram and can do maybe 1% more than a model that needs less than 1/10 of that. It's really close to car companies pushing everyone into SUVs when what we need are walkable cities, bikes, and microcars. Literally no one needs frontier models, but datacenter builds are basically making consumer silicon inaccessible. It's the same strategy as bulldozing neighborhoods and building highways through them so you can't walk or bike. It's a resource denial to force you behind their unnecessary society-destroying paywall. The parallels just grow more and more the closer you look.