Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend

Judgment

Developers deciding whether to build voice agents on Block’s Dessa must weigh more natural, speech-only conversation against awkward outputs and access limited to a small group of AI safety researchers.

Basis

Block announced Dessa, an AI model trained “from scratch on nothing but human conversation.” Dessa lead Ragavan Thurairatnam shared a sample chat where the model mistook the speaker for someone named “Ronaldo”; The Verge also describes demos about Christopher Nolan films, a surprise Paris trip, and a comparison with OpenAI’s GPT-Live-1 voice model. Block’s team says Dessa is a “true speech-to-speech model,” that “everything it knows came from human speech,” and that it can run locally on a MacBook Pro M4 Max “or better.” The stated reasons are curiosity, better understanding between humans and AI, and safety; the team argues voice AI that flattens speech into text misses human timing and emotion. Early access goes to “a small group of AI safety researchers.”