Overview
Jack Dorsey’s Block announced an AI model called Dessa that is trained entirely on human conversation, showcasing dialogue samples that exhibit natural yet sometimes uncanny interactions.
Block Introduces Conversational AI Model Dessa
You have got to listen to these conversations with Block’s new Dessa AI model.
Take this one, for example, shared by Dessa lead Ragavan Thurairatnam. The conversation between the model and the speaker does have a mostly natural back and forth.
Features and Examples
Demonstration examples on the project website include varied topics such as discussions on film directors, surprise trip announcements, and comparisons with other voice models, highlighting its unique cadence and conversational tangents.
Technical Specifications
According to the development team, Dessa functions as a true speech-to-speech model where all knowledge is derived from human speech. It is capable of running locally on hardware such as a MacBook Pro equipped with an M4 Max chip or higher.
The creators explain that the project was motivated by curiosity, improving human-AI communication, and addressing safety concerns related to voice inflection and emotional awareness in artificial intelligence interfaces.
Availability
The Dessa team stated plans to provide an initial release of the model to a select group of AI safety researchers for early evaluation.
