Jack Dorsey's new AI talks like a baby learning words
The company says the system runs on high-end Macs and aims to improve how artificial intelligence processes human emotion.
By Thiago Kovalenko ·
Block, the technology firm led by Jack Dorsey, released an artificial intelligence model named Dessa. The company states that the system was trained entirely on human conversation without any written text. To demonstrate the technology, Block published audio recordings of interactions between the model and various users. Some of these exchanges appear natural, while others sound slightly awkward.
Ragavan Thurairatnam, who leads the project, shared one specific sample. In that recording, the model initially mistakes the speaker for a person named Ronaldo. The Dessa website also features other examples. One clip discusses Christopher Nolan films, but the model shifts the topic to Bryan Singer's Superman movies. Another recording involves a user describing a surprise trip to Paris. The website notes that the model responds with what sounds like emotional gratitude. A separate clip pairs Dessa with OpenAI's GPT-Live-1 voice model, and that interaction sounds smoother.
The development team describes the system as a true speech-to-speech model. They claim that everything the model knows comes directly from human speech. According to the team, the software can run locally on a MacBook Pro equipped with an M4 Max chip or better.
The developers list three reasons for building the model: curiosity, improved understanding between humans and AI, and safety. They compare the training process to how babies learn to talk without seeing written words. The team notes that achieving this with a machine is difficult. They argue that current voice assistants reduce spoken language to text and fail to match human timing. This approach, they say, wastes the potential of speech as a natural interface.
The safety argument focuses on emotion. The team says that AI systems are becoming more independent but remain deaf to emotional cues in voices. They describe a scenario where a person in serious pain talks to a voice assistant that does not realize the urgency. The developers describe Dessa as a step toward fixing this issue.
Block plans to provide early access to the model for a small group of AI safety researchers. The announcement did not include a schedule for a wider public release.
