Jack Dorsey’s Block has announced Dessa, a new AI model that the company says was trained “from scratch on nothing but human conversation.” The Jack Dorsey AI model is built to talk by voice rather than text, and Block has shared sample chats to show how it responds. The model is not open to the public yet: early access will go to a small group of AI safety researchers.
What the sample chats show
Block shared conversations that Dessa lead Ragavan Thurairatnam had with the model. In one, the exchange is mostly natural, although the model first mistakes the speaker for someone called “Ronaldo.” Some of the other examples are awkward or uncanny, which makes the demos more revealing than a polished product tour.
The Dessa website carries more examples. One covers Christopher Nolan’s films and includes an unexpected tangent about Bryan Singer’s Superman films. In another, a person says they took the model on a surprise trip to Paris, and the site describes the reply as sounding like “teary gratitude.” A third pairs Dessa with OpenAI’s GPT-Live-1 voice model, and in the clips shown, Dessa comes across as the more natural of the two.
How the model was built
The Jack Dorsey AI model is described on Block’s website as a “true speech-to-speech model,” meaning it takes spoken input and produces spoken output. The team says everything Dessa knows came from human speech. According to the site, it can run locally on a MacBook Pro M4 Max “or better.”
The team compares the approach to how babies learn to talk before they ever read. It says the question was whether a model could learn the same way. Its answer is that one can, though it is very hard to do.
Why Block says it built Dessa
The team gives three reasons: curiosity, which it calls fun; better understanding between humans and AI; and safety. On safety, it argues that AI keeps gaining power while still missing the emotion in people’s voices. It gives the example of a person in serious pain whom a voice AI fails to notice, and presents Dessa as a step toward fixing that gap.
Block has not said when, or whether, Dessa will reach the general public. For now, the only group named for access is a small set of AI safety researchers.
Aasan alfaaz mein
Jack Dorsey ki company Block ne Dessa naam ka AI model dikhaya hai jo sirf insaani baat cheet se seekha gaya hai. Ye awaz mein seedha baat karta hai, aur kuch examples ajeeb aur bilkul insaan jaisay lagte hain. Abhi ye sirf kuch AI safety researchers ko early access mein milega, is liye aam log ise istemal nahi kar sakte. Company ka kehna hai ke ye voice AI ko jazbaat samajhne mein behtar banane ki taraf ek qadam hai.
