Replying to @⁨alakey@piefed.social⁩

What they are choosing is to not produce anything for the consumer market any longer.

They see these big companies trying to corner the market on cloud services and AI and they’re betting the future is letting these guys control everything.

If all consumers’ money has to be funneled through like 3 or 4 providers, best to bet on those providers and their subcontractors.

This is why capitalism sucks ass.

Replying to @⁨mereo@piefed.ca⁩

We don’t know if they’re actually more efficient because ClosedAI and Anthropic won’t tell us how (in)efficient their models are, whether they’re running quants, etc. Kimi K3 is about 3 terabytes of VRAM unquantized so I wouldn’t call the Chinese models super efficient either. The “efficient” models are the distillations and particularly quantized versions of some Chinese models. Nothing stopping ClosedAI and Anthropic from doing the same if they’re willing to lose a bit in accuracy.

Replying to @⁨SpruceBringsteen@lemmy.world⁩

Yeah this is like when the cops investigated themselves and find no wrongdoing. Why would anyone listen to the chip mafia when they have only their own interests in mind. They want you to think that there is no point in waiting for prices to drop, when that is obviously what everyone should be doing right now. Basically everyone should boycott the entire semiconductor industry as much as possible.

Replying to @⁨gedaliyah@lemmy.world⁩

thats what the economics textbook says, but it also says that all parties in the economy are perfectly rational. however this is not the reality. they have tested it, people are already used to the high prices and they have no problem with it, enough of them will buy it anyway. the price drops from completely unaffordable 7x price to barely affordable 4x price, and people are already buying it again. if they aim for 4x price, they only need the quarter of the former sales to make the same profit. and if at 4x price the sales are the quarter of former sales, they could theoretically shut down most factories or production lines to reduce costs and in turn raise profits even more.

Replying to @⁨anon_8675309@lemmy.world⁩

Self hosting also requires a ton of RAM because each individual user has to load a (usually slightly smaller) model + KV cache, whereas for hosted, they load the huge-ass model once (per cluster) and then store KV caches for a bunch of users, which has better scaling.

The memory companies win whether you self-host or pay a subscription. They might win more if you self-host because look at the cost of self-hosting, it’s actually pretty steep if you don’t have the hardware already from the before times.