News for hackers. Stuff that matters.

A comment on DeepSeek-v4.1-Flash-1017. Parent comment

Re:DeepSeek-v4.1-Flash-1017 by ReptileMan on Saturday October 17, 2026 @05:23AM (#50129366)
We have the weights, we have the papers, nothing preventing development on this foundation.
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by hgoel on Saturday October 17, 2026 @09:08AM (#50130840)
Except compute
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by ReptileMan on Saturday October 17, 2026 @09:29AM (#50131017)
Do you think we are near the theoretical maximums of how much model we cram into 20B?
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by hgoel on Saturday October 17, 2026 @11:17AM (#50131876)
I don't consider myself qualified to make that call, though I believe there's plenty of room for improvement. But I don't mean inference compute, which is fairly accessible for well-off enthusiasts even now. Training is a lot more expensive though.
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by ReptileMan on Saturday October 17, 2026 @03:54PM (#50134483)
Back in the day we had seti@home, prime95 and fold@home. We will make it work.
[ Parent | on HN ]
Re:DeepSeek-v4.1-Flash-1017 by Azantys on Sunday October 18, 2026 @06:15AM (#50139662)
When models are in the hundreds of gigabyte range, there are not enough nodes to train, because most people dont have 100k in AI hardware. Also there is no training method for machines scattered around the world as far as I know.
[ Parent | on HN ]