This AI learned to talk without ever reading a single word.
Block’s research lab unveiled Dessa, a voice model trained entirely on human conversation — no text, no pretrained models — that runs locally on a MacBook Pro M4 Max.
Dessa is a full-duplex voice model, trained from scratch on speech alone, on a single NVIDIA GB200 rack.
It learned emotion, sarcasm, and laughter natively, the stuff text-trained models fake, and it runs locally.
It’s not public: research preview only, a small group of safety researchers, no benchmarks, just demo recordings. One demo even mistook its speaker for Ronaldo. So the voice is human. The question is whether we’ll still be able to tell.
Watch the reel: https://www.instagram.com/p/De71JjrgvVx/
Sources: TechCrunch (“Block’s new voice model sounds eerily human”), Block research lab. Instagram: https://www.instagram.com/simpleaiexplains/ Support: https://buymeacoffee.com/simpleaiexplains
Leave a Reply