So DeepSeek situation summarized:
*They are not a small engineer team but one of the leading frontier lab (+100 researchers full time).
*They are not a newcomer. Started in 2023 by retraining a llama, then slowly rising to the top. All documented in their 16 (!) papers.
Indianapolis residents have shut down a proposed Google data center.
The tech giant wanted to build a massive 500 acre facility, but people got organized and stopped them.
The $1 billion data center would've used one million gallons of water a day.
I feel this should be a much bigger story: DeepSeek has trained on Nvidia H800 but is running inference on the new home Chinese chips made by Huawei, the 910C.
Maintenant que les comptes Twitter Blue peuvent poster des vidéos d’une heure (et que la modération est à la ramasse), Twitter est en train de devenir un Pirate Bay mainstream : 6 millions de vues pour une copie piratée du film Mario Bros.
So it seems we may finally have a GPT-4 level model in open source. reddit.com/r/LocalLLaMA/c… It's a merge of two llama 70b and since we live in the best AI timeline it's created by an anon with an avatar that looks like this:
Des personnalités "de gauche" quand Trump perd son twitter après avoir soutenu une insurrrection armée :
▶ 🔘──────── 20:13:67:23
Les mêmes quand des féministes sont suspendues en masse pour ouvrir un débat de fond.
▶ 🔘──────── 00:02