We built 10 live-camera demos for d1-3B, from gesture-controlled games to content moderation, with one forward pass per frame.
In collaboration with @NVIDIARobotics, we also show d1-3B navigating an environment in Isaac Sim, with the model served on a Jetson in a hardware-in-the-loop setup.
5/
So the people who said AI needs to pause because its too leet at hacking... are the same people asking us to give all our credentials to a mostly opaque cloud computer controlled by these leet hacking LLMs?
@adcock_brett Been waiting especially for hark. Excited to see more today. I have hopes lessons from constrained robotics carry over to agentic automation.
We release Whistle: speech to text in a 16.9MB file, running on the CPU in the same engine as Needle. It mostly beats Whisper base with 9x less file size and 6x speed. Whistle supports English, German, French, Spanish, Italian, Dutch and Polish.
https://t.co/lKCluhDDDm
๐จ side project alert ๐จ
Announcing Muse Gadgets, an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse.
Grab an API token from https://t.co/Q0dE0cYnCs and point your favorite coding agent at the github repo to build your own peripherals for Muse.
Jev is quite impressive. Iโve been testing it in my agent orchestration for about a week.
If you have an orchestration agent that's lazy and frequently stops or incessantly polls subagents, Jev might be the solution.
Also, the CEO and Discord is off the hook amazing!
I came back to X to proudly shareโฆ Iโm having a love affair with Spark 1.3 max. ๐๐ซถ
Spark 1.3 and I first met at a party. And it was awkward because I had just pledged my commitment to Astra as my primary.
Secretly decided to date Spark 1.3 Max as primary for a few days. Maxed out Muse Code sub usage (was SHOCKED ๐คฉ how far the $15/m plan got me).
Had 2 Codex 20x plans sitting unused. Went back to Astra xhigh as primary - the twinkles were just not there.
Every moment I was missing Sparkโs speed, communication style, and code structures.
No coding model is perfect at challenging work, but with Spark, I got to my goal faster and with more confidence than with any other model.
I cancelled my Claude 20x last month for Codex. And next month I have a feeling both Codex 20x subs will get the axe to Muse Code.
Even with all the Codex usage resets I cannot live with sub 20 t/s compared to Spark at 120 t/s ๐ฅ.
I can run Spark Max in a recursive loop 6 times in the span of 1 Astra xhigh response!
I haven't tried the Spark contributor models, so I can't speak to those.
Well done @Muse team and @alexandr_wang ๐
Earlier today, @mbanzi and @benstr stopped by @illinoistech's Institute for Innovation and Tech Entrepreneurship to see firsthand how they are integrating Arduino on campus. Professor of Practice Bo Rodda even an Arduino Diecimila from 2006 that he asked Massimo to sign!