Block, the company led by Jack Dorsey, has announced a voice AI model called Dessa. Block says it was trained "from scratch on nothing but human conversation." The demo chats are mostly natural, but they can also get a little awkward and uncanny.
What Dessa does
Dessa is described as a true speech-to-speech model. It works with spoken sound directly. Block says many voice AI tools flatten speech into text first, which loses timing and tone. Dessa can also run on a laptop, specifically a MacBook Pro M4 Max "or better."
The demos show the quirks. In one, Dessa plays a guessing game but first mistakes the speaker for someone named "Ronaldo." In another, it veers into Bryan Singer's Superman films during a talk about Christopher Nolan. When a person describes a surprise trip to Paris, Dessa responds with what sounds like teary gratitude.
Why Block built it
The team gives three reasons: curiosity, better understanding between humans and AI, and safety. Its worry is that AI is nearly deaf to emotion in our voices, such as a person in serious pain being ignored. A small group of AI safety researchers will get early access.
What is not known yet
The material does not say when the public can try Dessa, what it will cost, or whether it will show up in any Block products. The examples come from Block itself, so independent testing is still to come.
The takeawayDessa is an early, company-shown demo of a voice AI that aims to hear emotion, and it is not yet something the public can try.