Jack Dorsey's Block unveils Dessa, a voice AI model trained on human conversation

Block, the company led by Jack Dorsey, has announced an artificial intelligence model called Dessa, which it says was trained "from scratch" on nothing but human conversation. To show what the model can do, the company shared recordings of chats with people. As technology outlet The Verge notes, some of them come across as slightly awkward and uncanny.

In one example shared by Dessa lead Ragavan Thurairatnam, the exchange is mostly natural, although the model at first mistakes the speaker for someone named "Ronaldo." A video also shows the model playing a guessing game by voice.

Other examples

The Dessa website features several more samples. In one conversation about director Christopher Nolan's films, the model goes off on an unexpected tangent about Bryan Singer's Superman films. In another, a person tells Dessa about taking it on a surprise trip to Paris, and the website says the model answers with what sounds like teary gratitude. There is also a conversation between OpenAI's GPT-Live-1 voice model and Dessa, in which Dessa sounds more natural.

What the developers say

The Dessa team describes it as a true speech-to-speech model, with everything it knows coming from human speech. According to the team, it can run locally on a MacBook Pro M4 Max or better.

The developers give three reasons for building it: curiosity, better understanding between humans and AI, and safety. They note that babies learn to talk without ever seeing a written word and say they wanted to test whether a model could do the same. In their view, today's voice AI flattens speech into text, cannot keep up with human timing, and remains nearly deaf to emotion in people's voices. They describe Dessa as a step toward fixing that.

The team says it will give a small group of AI safety researchers early access to the model.