Bros, breaking news. Su Zifeng brought a host computer the size of a lunch box to the stage and ran a…
Bros, breaking news. Su Zifeng brought a host computer the size of a lunch box to the stage and ran a large model with 235 billion parameters on the spot. It doesn’t rely on the cloud or rents a GPU, it relies entirely on AMD Ryzen AI Max+ 395, the world’s first x86 processor with 128GB of unified memory where the CPU and GPU share unified memory. Let’s compare: the RTX 5090 only has 32GB of video memory, the RTX…
Deepseek V4 Flash running locally on a Mac Studio is too honest.
🤣 "What model are you?"" I am Claude "
Everyone is sleeping on the cheapest AI trick nobody talks about
Moon dev: "i was sitting there watching deepseek take fifty whole seconds and i just snapped" "cheap means nothing if it's quietly stealing hours of your life and nobody warns you about that"
Cycling Pelican Test
DeepSeek V4 Flash (local 2bit) vs Qwen3.8-Flash-Next mixed 4/8bit vs Claude Sonnet 5 3 times each under the same conditions. (inference on, temp 0.3, 28k tokens) Result: DeepSeek: Simple line art. shortest and lightest Qwen mixed 4/8bit: Color cartoon. The most vibrant and highly complete…
You have reached the end of the archive
All of deepseek