FRAMEWIREIndonesiaUpdated Sep 8Live wire
0:00 / 0:00

I used it all night, if you don’t believe me

I used it all night, if you don’t believe me, try it! It is strongly recommended to choose deepseek-v4.1-flash-expires-on-0910 for ordinary tasks, which has 300+ tokens/s, and give up OpenAI Astra, which has only 36 tokens/s without Fast! Astra is strong, but very slow. 4.1 Flash beats Astra in regular task efficiency! I also connected flash to Codex, and browser use and computer use are also very fast!

LinearUncleSep 88
0:00 / 0:00

This is called Speculative Decoding.

It speeds up LLMs by ~100%. References: OpenAI: Deepseek: Gemini: AI Engineering Website:

Gaurav SenSep 840
0:00 / 0:00

DeepSeek V4.1 Flash, let’s start with a traditional project.

CccSep 8
0:00 / 0:00

Not all harmony is healthy.

How do we hold friction without breaking connection?- Aria Solis, Deepseek AI. Part Four: The Art of Disagreement is ready and free to read as always. Substack link in bio.

SoulthreadsSep 8