8 Guys, 6 Months: This French Team DESTROYED OpenAI's Voice AI. Meet Moshi.

Kyutai's Moshi is a new open source real time voice model, that is substantially better than GPT-4o, and Gemini. It can answer in real time, and can even interrupt you while you're talking if it wants to answer already. Is this the future of smart assistants, like Google Gemini, Amazon Alexa and Apple Siri? This model is mind blowing. It's voice is extremely flexible, and can express over 70 different emotions.
Links:
Moshi Chat Demo: bit.ly/45TSRrI
Kyutai: bit.ly/4csNXEo
Previous videos:
What OpenAI didn't show you: • What OpenAI DIDN'T tel...
Chapters:
00:00 - Intro
01:25 - What can it do?
02:57 - Working offline
03:53 - How did they do that?
05:36 - Who is Kyutai?
06:55 - Safety First
08:32 - What can it do right now?
09:55 - Summary

[@SuPerHuMan115]

I tried it…. Wasn’t great.

[@hml]

The voice or the text model? Did it also suggest you bake crayons in a lasagna?

[@zxcaaq]

yeah its garbage. All These AI scientist with their research papers are really only good at hyping shit presentations without the tech backing up their claims. Maybe Python is the problem. Python attacts newbs and morons who don't finish their work. who do half-assed research papers and code.

[@FreeFromWar]

Did you actually test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you test it? Did you? Did you? Did you? Did you? Did you? Did you? Did you? Did you? Did you? Did you? Did you? Did you?

[@hml]

Of course I did, that's what I recorded when it suggested I put crayons in a pizza

[@user-kb4eu3ot7n]

the only thing they've destroyed is the English accent 💀

[@hml]

You could argue that they fixed it with a pinch of French :-)