Summary
Block, the company led by Jack Dorsey and best known for Square and Cash App, has unveiled Desa, an AI voice model that many listeners say is nearly impossible to distinguish from a real person. This video explains what Block built, why it sounds so natural, and why the release has drawn both excitement and concern online.
Desa is described as a full duplex speech-to-speech conversation model. Unlike typical voice assistants that convert speech to text, run a language model, and then synthesize a reply, Desa takes in audio and produces audio directly. That lets it listen and talk at the same time, laugh mid-sentence, interject with small acknowledgments, and interrupt the way people do. Block says it was trained from scratch only on recordings of human conversation, with no text and no existing language model. The video walks through why that training choice captures the pauses, stumbles and emotion that written text leaves out.
The second half focuses on risks, especially scam calls. The video considers how a convincing real-time voice combined with voice cloning could scale family-emergency and fake-bank scams, and why detection tools and watermarking face an ongoing arms race. It also notes that Block has not released Desa publicly, offering a waitlist and early access for AI safety researchers.
📺 Source: TheAIGRID · Published October 25, 2026
🏷️ Format: News Analysis
