Block, the company led by Jack Dorsey, announced an AI model named Dessa, described as a full-duplex speech-to-speech system trained “from scratch on nothing but human conversation.” According to the company, it is positioned as a model you can have a realistic spoken conversation with, trained entirely on speech.
To demonstrate the model, Block shared sample conversations between Dessa and human speakers. Dessa lead Ragavan Thurairatnam posted one such exchange, which reportedly featured a mostly natural back-and-forth but also moments that were awkward or uncanny — including an instance where the model initially mistook the speaker for someone named “Ronaldo.”
Why it matters
The announcement highlights an approach to conversational AI built directly on speech rather than text, and the shared examples illustrate both the potential for natural dialogue and the current rough edges.
Who should care
Developers and companies interested in voice-based AI interfaces may want to watch how Dessa’s speech-to-speech approach develops.