RT Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)Re my point is: if they can train a 1.6T 48B active on 910Cs using all these tricks (that's the…
aside from the obvious sneering about chip provenance,> Meanwhile, the largest U.S. clusters are now approaching 1 million H100e: https://epoch.ai/data/ai-data-centersI wonder, what do the AI policy…
Xiaomi MiMo code repo includes system prompts for a number of model families and regimes, including "beast.txt" built for an *extremely* persistent autonomous /goal-chasing agent. "MUST…
Google remains very powerful in image/videogenLogan Kilpatrick: Introducing Nano Banana 2 Lite 🍌 and Gemini Omni Flash 🔮, our new generative media models in the Gemini…
zAI denies the rumor about operating a router to Claude(a more specific rumor but I'll take it as a blanket denial)people are free to decide how…
«I find it hilarious that people think training an AI model on chips is an accomplishment... that's what chips are supposed to do! obviously…»First principles thinker,…
Taste PaysI hope they expand their compute. K3 will probably be sensational. Definitely a lab that makes a unique contribution to the entire industry.Poe Zhao: Moonshot…
insanethis is the sort of thing you should be doing if you can't just rely on Moore to bail you outEtched: We're scaling production as fast…
wow did they discover speculative decoding or something?margins go up again!Stephanie Palazzolo: OpenAI engineers earlier this month developed an optimization that cut inference costs in half…
RT Nathan LambertWhen we were in China, @xeophon and I made a quick detour to visit Meituan. They continue to be one of our favorite open…