DeepSeek is an AI assistant which appears to have fared very well in tests against some more established AI models developed in the US, causing alarm in some areas over not just how advanced it is, but how quickly and cost effectively it was produced.
[…]
Individual companies from within the American stock markets have been even harder-hit by sell-offs in pre-market trading, with Microsoft down more than six per cent, Amazon more than five per cent lower and Nvidia down more than 12 per cent.
Smaller and more local is how this was always bound to go.
Mainframes have never been a good idea. The only sensible applications for computers the size of a building are where one task needs the whole computer. Accessing a slice of that, from a lesser machine, is rarely more useful than letting the lesser machine flex whatever power it has. Almost never, if they can all access a dumb server.
LLMs started on consumer video hardware. The giants in this specific field went big because it was the only advantage they could press. They knew, once it shifted down to more-efficient models that any desktop or phone could run… they were fucked. And here we are.
It won’t go much better for DeepSeek, either. They’ve fucked themselves by implementing the blatant censorship everyone expects and laughs at - so right out the gate, people are modifying their models. There’s not going to be any one country or company that supplies a ready-to-go thingamajig that everybody uses. It’s gonna stay weird and democratized for a good long while.