Block's new AI learned to talk just by listening, with no text at all
Jack Dorsey's company Block says its Dessa voice model was trained only on human conversation. It laughs mid-sentence and gets louder when you ask it to speak up.
Most voice assistants work like a translator. They turn your voice into text, think in text, then read an answer back out loud. Block, the company led by Jack Dorsey, says that throws a lot away. Its small research lab built Dessa, a voice model that skips text completely.
The team says it trained Dessa "from scratch on nothing but human conversation." No text, and no ready-made model to start from. It learned to talk simply by listening. As far as Block knows, this is the first AI you can have a realistic live chat with that learned only from human speech.
In the demo recordings, Dessa listens and talks at the same time, like people do. It says "yeah" or "got it" while you speak. It laughs in the middle of a sentence when something is funny. When a person asks it to speak up, it gets louder. The team writes: "Nobody taught it that."
Block says Dessa learned language from hundreds of times fewer words than text-based AI needs. It can even run on a high-end MacBook Pro laptop.
The team says this is also about safety. It writes that voice AI is still "nearly deaf to the emotion in our voices." Its example is a person in serious pain while a voice AI has no idea and just keeps going. The team calls Dessa a step toward fixing that.
For now, Dessa is a research preview. Block is giving early access to a small group of safety researchers, and launch links are "coming soon."
Open it and check the claim yourself. That is the point.