Jack Dorsey's company Block announced the release of Dessa, an artificial intelligence model designed to hold realistic conversations using speech. The technology is a full-duplex speech-to-speech system, meaning it processes audio input and generates audio output simultaneously rather than through text intermediaries. Block stated that the model was trained from scratch on human conversation data, distinguishing it from systems that rely on text-based training methods.
To demonstrate the model's capabilities, Block shared examples of chats between Dessa and human speakers. Ragavan Thurairatnam, the lead on the project, posted one such exchange online. In this specific interaction, the dialogue maintains a mostly natural back-and-forth rhythm. However, the conversation also highlights moments where the model's responses appear awkward or uncanny to listeners.
One notable error in the shared example involves the model initially mistaking the human speaker for someone named Ronaldo. Despite this glitch, the presentation aims to showcase that Dessa can sustain a realistic conversational flow. The announcement positions the new tool as the first AI model of its kind to be trained entirely on speech data for the purpose of direct human interaction.