"We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built," Dorsey wrote. "Talking with it feels like talking with a person."
Big claim. So I went to the model's website and listened to every clip on it.
A model that learned like a baby
Dessa is a true speech-to-speech model. Audio goes in, audio comes out, and at no point does anything get flattened into text. That is the opposite of how today's voice AI works, and the Dessa team thinks that is exactly the problem.
The lab behind it is a small AI research team at Block led by Ragavan Thurairatnam, founded by former members of Dessa, a Toronto deep learning startup. They say they trained the model from scratch on nothing but human conversation. No books, no web pages, no pretrained models. Like a baby, it learned to talk only by listening.
It is also full-duplex, meaning it listens and talks at the same time, the way you and I do. Inference runs faster than real time, so there is no awkward wait, and the whole thing runs locally on a MacBook Pro M4 Max or better.
The team gives three reasons for building it. Curiosity, better understanding between humans and AI, and safety.
"Imagine being in serious pain while a voice AI has no idea and just keeps going," the team explains. "That's the safety problem. Dessa is a step toward fixing it."
I listened to the clips. It is different.
The Dessa page carries live recordings of real people talking to the model. Nobody steers it, the team says. Dessa picks what to say, when to say it and how, right down to its own voice and personality.
In one clip, a man tells Dessa about his day, and it laughs the moment something is funny, mid-sentence, the way people do. Then it does something I have never heard an AI do. It tells him he sounds far from the mic and asks him to come closer so it can hear him better.
Hear the release-day conversation on the Dessa page.
In another, a caller surprises Dessa with an anniversary trip to Paris. It gasps "oh my god" under its breath, then answers with what sounds like teary gratitude, before joking that a French breakfast is an espresso and a cigarette.
Hear the anniversary surprise clip on the Dessa page.
My favorite is the one where Dessa gets it wrong. It greets a man named Dheer as "Ronaldo" and has to be corrected.
"I'm not Ronaldo," the caller laughs. "You're mixing me up with another guy."
A synthetic voice making a very human mistake is, weirdly, the most convincing thing on the page.
Then there is the clip where Dessa talks to OpenAI's GPT-Live-1 voice model. This is my subjective read, but the gap is hard to unhear. OpenAI's model sounds like a polite assistant reading from a card. Dessa sounds like the friend who set up the call.
Hear Dessa talk to GPT-Live-1 on the Dessa page.
Why this matters
Here is my honest opinion. Voice is the last interface that actually matters, because it is the only one we were born with. We did not evolve to type. We evolved to talk.
And so much of daily life still runs on phone calls. Restaurant bookings, customer support, doctor's appointments. My restaurant test is the only benchmark that counts, and the first voice AI that a restaurant owner cannot clock as a machine changes who answers the phone everywhere.
That cuts both ways, and I would be lying if I said it does not worry me. A voice that passes for human is also a voice that can scam someone you love. Block says it is giving a small group of AI safety researchers early access before any wider release, with more recordings coming soon. Given what is at stake, that is the right order.
For now, Dessa is not something you can try yourself. You can only listen, and I would tell anyone who cares about where AI is going to spend ten minutes on its website doing exactly that.
The bitcoin billionaire behind it
There is one more reason this launch does not surprise me, and it has nothing to do with AI.
Dorsey has spent years as one of Bitcoin's loudest backers. Block's bitcoin revenue hit $2.42 billion in a single quarter, TheStreet reported. And he told the Presidio Bitcoin podcast that bitcoin fails through irrelevance if it stays a store of value and nothing more.
"I think it fails through irrelevance," he said. "It fails to be relevant to people on a daily basis."
That is the same instinct behind Dessa. Technology only counts when ordinary people actually use it. Bitcoin has to be spent. Voice AI has to pass for a person on the phone. Say what you want about Jack Dorsey, he keeps betting on the same idea. The future has to sound, feel and work like something a normal person would actually touch.
I have heard every big voice demo of the past few years, and nearly all of them still failed my dead-person test. Dessa does not. For the first time, I heard a machine and forgot, just for a second, that it was one.
thestreet.com
