Block's Dessa Learns Language From Spoken Conversation, Not Text

Block, the payments company led by Jack Dorsey, has released Dessa, an AI model it says was trained "from scratch on nothing but human conversation." The model converts speech directly into speech rather than passing audio through written text.

The company frames the approach as an attempt to reproduce how humans acquire language. "Babies learn to talk without ever seeing a word written down," the Dessa team wrote, adding that replicating the process for a model "is just very hard." Block lists curiosity, better understanding between humans and AI, and safety as its three motivations for building it.

Block argues that converting speech to text before processing it degrades the experience. The Dessa team contends that speech is the most natural interface to any intelligence and that current voice AI "flattens what we say into text and can't keep up with human timing." The safety rationale rests on tone: the team writes that AI systems grow more capable while remaining "nearly deaf to the emotion in our voices," and points to a scenario in which a person in serious pain talks to a voice assistant that does not recognize distress. Dessa is described as "a step toward fixing" that gap.

Block published recordings of Dessa talking with people. Ragavan Thurairatnam, who leads the Dessa project, shared one exchange in which the model initially mistakes the speaker for someone named "Ronaldo." On the Dessa site, a conversation about Christopher Nolan's films drifts into a discussion of Bryan Singer's Superman movies, and another clip has Dessa responding to news of a surprise trip to Paris with what the site calls teary gratitude. A recording pairing Dessa with OpenAI's GPT-Live-1 voice model sounds more natural than the other examples.

The Dessa team says the system is a "true speech-to-speech model, and everything it knows came from human speech." It runs on a MacBook Pro M4 Max "or better." Block plans to give "a small group of AI safety researchers" early access, and the announcement did not include a wider release schedule.