Block, the financial services and digital payments company led by Jack Dorsey, has introduced Dessa, a new artificial intelligence model developed from speech-based training data. Unlike conventional language models that convert speech to text before processing, Dessa operates as a direct speech-to-speech system, learning entirely from patterns in human conversation without relying on written language.
The company released several demonstration conversations showcasing Dessa’s conversational abilities. Examples include discussions about filmmaker Christopher Nolan’s work, a dialogue about a surprise trip to Paris, and a voice-based guessing game. In demonstrations, the model occasionally exhibits natural conversational flow while sometimes producing responses that feel stilted or contain minor errors, such as initially misidentifying a conversation partner.
According to the Dessa development team, the model was designed with three primary objectives: exploring the feasibility of speech-based learning, improving human-AI interaction through natural dialogue, and addressing safety considerations. The team emphasized that the approach mimics how infants acquire language without exposure to written text. They also argued that speech represents the most natural interface for human-AI communication and that current voice systems lose important emotional and timing information by converting speech to text.
The team noted a potential safety advantage: since Dessa processes speech directly, it theoretically can better recognize emotional context, such as distress in a user’s voice, rather than processing only the literal text of what someone says. The model is capable of running locally on higher-end Apple computers, specifically a MacBook Pro M4 Max or equivalent hardware.
Block stated it will provide early access to Dessa to a select group of artificial intelligence safety researchers for further evaluation and testing.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI