Block releases AI that learns to talk without reading
The company behind Square says its new model, Dessa, was trained exclusively on spoken human interaction and can run on a single laptop.
By Luana Sato ·
Jack Dorsey’s company Block has released an artificial intelligence model that processes only spoken language. The system, named Dessa, was developed without any text data.
The team states the model was built from scratch using nothing but human conversation. They claim it is a true speech-to-speech model. The developers say it can operate locally on a MacBook Pro M4 Max or a more powerful machine.
Ragavan Thurairatnam leads the project. He shared a sample exchange where the model initially mistakes the speaker for a person named Ronaldo. The conversation quickly corrects itself. Other examples on the project website show varied results. In one clip, the model shifts from discussing Christopher Nolan films to Bryan Singer’s Superman movies. In another, a user describes a surprise trip to Paris, and the model responds with what the site describes as teary gratitude.
The developers argue that current voice assistants fail to capture the full value of speech. They say existing systems flatten spoken words into text and cannot match human timing. They view speech as the most natural interface for intelligence.
The team lists safety as a primary motivation for the project. They note that AI systems are gaining independence but remain largely unaware of the emotion in human voices. They describe a scenario where a person in severe pain speaks to a voice AI that continues without noticing the distress. They call Dessa a step toward addressing that gap.
Block plans to give early access to a small group of AI safety researchers. No wider release schedule was provided in the announcement.
The developers compare the approach to how babies learn to talk. They note that infants learn speech without ever seeing written words. They state that while it is very hard, the model proved it could do the same.
