Cast by purp
Show the cast this replies toYeah, from how ive read it is, basically everyone internally knows they are two years behind, but they also are where they can atleast train new models. Im assuming they are going to use extra capacity as backup/fill in the gaps where their chips are still short( programming language and scope of chips needed to train a model, 160k for deepseek v.5)
0 replies
No replies yet