Bluesky gets the lion’s share of press when it comes to Jack Dorsey’s non-Twitter projects, so it might be easy to forget he was also one of the developers of Square, the ubiquitous point-of-sale system that has a number of noteworthy subsidiaries like Cash App, Afterpay, and Tidal.
Square also changed its name to Block, Inc. in 2021. Perhaps sensing a way to acquaint the unfamiliar with the rebranded business by way of earned media hits, Block just released an AI model with a bit of a twist.
The model’s name is Dessa and Block’s announcement post says it was “trained from scratch on nothing but human conversation.” Block claims that “talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
As reported in The Verge, Dessa processes and generates audio directly using speech-to speech architecture rather than converting input speech to text and generating a response like conventional voice models. This duplex architecture also allows Dessa to speak and listen at the same time, which should prevent some of the abrupt subservient silences most other AIs offer the moment a user pipes up.
Block understands that, for things like these, hearing is believing, so the company put out a number of audio examples of Dessa speaking to showcase its abilities.
Take this video on X, posted by Dessa team lead Ragavan Thurairatnam, where a Block intern plays The Guessing Game with the model, having it guess the everyday items she’s thinking of. It’s kind of funny that Dessa inadvertently spoils things by immediately guessing the object on the first try, twice in a row. It’s even funnier when it shows some of the typical limitations of AI comprehension skills by carrying on with guessing after being told it won. But what stands out most is just how uncannily human Dessa’s voice does sound.
In another video post on X by Thurairatnam, a differently voiced Dessa speaks with the unmistakable cadence of an African American man. Asked how he’s doing, this Dessa replies, “man, I’m doing good, breh. I ain’t gonna even lie. I’m tired.” Dessa goes on to explain that his fatigue comes from having to go to work and pick produce. He then opines about his new baby boy, who he and his wife are calling Darius Johnson. Oddly the fictional newborn doesn’t seem to account for Dessa’s alleged sleepiness. You can likely guess the sort of replies this post got.
Some other videos and clips show the model’s range of intonations and emotions, like empathy and tearfulness, as well as a side-by-side with ChatGPT, which sounds downright robotic compared to Dessa’s naturalistic flow.
Block says the Dessa team has already granted “a small group of AI safety researchers” early access to the model, which can run locally on any MacBook Pro M4 Max “or better.”
Dessa’s creators say the model was made for the sake of “curiosity (aka fun), better understanding between humans and AI,” and to address the safety problems that arise from AI models being “nearly deaf to the emotion in our voices.”
We’ll have to wait and see what Block ultimately winds up doing with Dessa. But once a model this convincingly human is out in the wild, all of us humans are going to have to update our internal AI-clocking firmware.
