DeepSeek’s spending shows building AI doesn’t need billions

Watch DeepSeek’s spending shows building AI doesn’t need billions - Rizzle on Dailymotion

Transcript

00:00deep seek spending shows building ai doesn't need billions when deep seek launched its r1 model

00:06earlier this year it briefly sent shockwaves through silicon valley how did a relatively small

00:12chinese startup manage to build a competitive large language model on what looked like couch

00:17cushion money compared to open ai's billions now thanks to a new paper in nature we have the

00:22receipts 294 000 and 512 nvidia h 800 chips that's not pocket change but in ai terms it's basically a

00:32budget ramen diet compared to open ai's wagyu beef spending the secret sauce trial and error

00:38reinforcement learning deep seek's team avoided costly annotated data sets allowing the model to

00:44find correct answers independently like a child playing a video game collect gold coins good run

00:50into enemies bad excelling in math and programming it pursued high scores to effectively solve problems

00:56autonomously the trade-off r1 occasionally offers explanations longer than a game of thrones novel

01:03mixing chinese and english mid-thought more concerning it avoids coding for politically

01:08sensitive topics like tibet or taiwan generating less secure code it's a reminder that ai reflects

01:15the values and politics of whoever builds it for now deep seek's experiment shows there might be

01:20more efficient ways to train models than burning mountains of cash