Astra-6 vs DeepSeek-v4.1 cloud vs local.
Same water prompt. Left → right: • Astra: “Worked for 30s” in ChatGPT • DeepSeek V4.1 Cloud: 103s, 28.5K completion tokens • DeepSeek V4.1 Local: 4× DGX Spark, 25m to HTML through Hermes 512×512 mesh. Click-to-ripple. Interacting
OKAY - took a long time but finally got some decent results with Deepseek v4.1 flash and 3D generation!
This is... well, not as straightforward than with GTP6 Astra 😅 But absolutely UNBEATABLE unit economics. Courtyard model took 34min to create, for a total of... $0.2 ! 🔥
Kimi K2, GLM-5 and DeepSeek-V4 all trained with a Newton-Schulz-orthogonalized optimizer that Moonshot…
Kimi K2, GLM-5 and DeepSeek-V4 all trained with a Newton-Schulz-orthogonalized optimizer that Moonshot AI measured at roughly 2x AdamW's compute efficiency.
It was the morning that my resignation letter was read 150 million times. In the same week
It was the morning that my resignation letter was read 150 million times. In the same week, a model with a third of the unit price became an official version, and the photos are starting to come with proof of authenticity. The quickening side and the checking side moved in the same week. 1️⃣ Unit price 1/3, deadline is September 14th 2️⃣ Gemini resides on Windows 3️⃣ Prove it's not AI 4️⃣ The person inside got off.
You have reached the end of the archive
All of deepseek