DeepSeek V4.1 Flash reaches 1 token/second on MacBook with 24GB of memory.
This is an improvement over the original speed due to SSD streaming and optimizations to the inference engine. Local AI also feels like a creative experiment that explores the boundaries between hardware and software. via：@mfranz_on
Everyone is talking about Astra improving intelligence.
Ask a robot for the colour made by mixing yellow and blue
Then watch what it picks up this 2025 visit to DeepMind is worth watching for the actual demos. robot section starts around 7 minutes in my version of that test for DeepSeek is smaller: give it a broken interface and
Pastel style beatiful Sakura Island with a train track loop
Created by DeepSeek-v4.1-Flash One-shot, three.js, single html file, no external assets.
You have reached the end of the archive
All of deepseek