Reported by 2 outlets
The story so far
AI summary of the reporting below · Updated · How we work
Jack Dorsey's Block has announced Dessa, an artificial intelligence speech model trained entirely from scratch on audio recordings of human conversations. Unlike conventional assistants that rely on text processing or pretrained language models, Block says Dessa learns exclusively from speech. The company claims the system operates as a full-duplex model that can listen and speak simultaneously without noticeable pauses, enabling it to interrupt, acknowledge speakers, and react in real time. [1][2]
Block says this voice-only training allows Dessa to capture conversational subtleties such as hesitation, sarcasm, laughter, and shifts in emotional tone. In company demonstrations, the model even held an unprompted conversation with OpenAI's GPT-Live-1. Block claims it trained the model on a single Nvidia GB200 NVL72 system containing 72 Blackwell GPUs, and that the resulting software can run locally on an Apple MacBook Pro with an M4 Max chip. [1]
Dessa is not yet widely available to the public. Block is initially inviting a limited number of AI safety researchers to test the system, while other users must join a waitlist. The company has not released its parameter count, model weights, or standardized benchmark results. [1]
Block has unveiled Dessa, a new speech-to-speech AI model trained entirely on human conversations without written text or pretrained language models. The system supports natural, real-time conversations with interruptions and emotional cues, and reportedly runs locally on a MacBook Pro equipped with Apple's M4 Max chip.
via Notebookcheck
Jack Dorsey's Block announced an AI model called Dessa that's trained "from scratch on nothing but human conversation," and to show off the model, Block shared chats it has had with humans that can occasionally be a little awkward and uncanny. Take this one, for example, shared by Dessa lead Ragavan Thurairatnam. The conversation between…
via The Verge