Post #818628
2026-03-26 19:28 UTC
IMO the most likely cause of any kind of #AI bubble pop are Chinese models breaking through. And at least in the open inference markets, that appears to be happening already:
"Since February, Chinese AI models made by groups such as DeepSeek and MiniMax have overtaken US rivals in token consumption, according to OpenRouter data"
A true SOTA model from China could send markets for a tailspin just like the original DeepSeek release did.
https://www.ft.com/content/2567877b-9acc-4cf3-a9e5-5f46c1abd13e?syn-25a6b1a6=1
Replies (1)
-
@dryak@mstdn.science 2026-03-26 19:58
@eloquence@social.coop Specially since Deepseek's approach of publishing good research papers, releasing the model, and making smaller models that one can locally goes contrary to the trend in the USA to be very secretive about models themselves, and pretend the only way forward is crazy large models that must be run in bonkers expensive continent-spanning datacenters in a giant arms' race.