Dorsey's Block Launches Dessa — AI That Speaks Like a Human
Is a human-sounding voice AI a gift to scammers or the breakthrough that finally makes talking to machines feel natural?
Story
Above: Jack Dorsey speaks during the Bitcoin 2021 conference in Miami, Florida, on June 4, 2021. Image credit: Eva Marie Uzcategui/Bloomberg/Getty Images
The Facts
- Block, Jack Dorsey's payments company behind Square and Cash App, has unveiled an AI speech system called Dessa, which it says was trained entirely on human speech, from scratch, without any text data, by processing audio directly rather than first converting it to text.
- Dessa is a full-duplex system, meaning it can listen and speak at the same time, which Block says lets it handle interruptions, laughter and shifts in tone. In one clip, it supplies a movie title during a conversation about Christopher Nolan's films.
- Block said it trained Dessa on a single NVIDIA GB200 NVL72 computing system and that the system is small enough to run locally on a MacBook Pro with an M4 Max chip or better, with inference running faster than real time.
Sources Split
LEFT
CENTER
RIGHT
+2
+2
+2
+1
ESTAB
CRIT
CENTER
PRO
ESTAB
The Spin
Industry-critical narrative
A voice this convincing is a gift to scammers, and the fallout lands on anyone who answers a phone. Once a machine fakes laughter, hesitation and warmth, the old gut check for spotting a fake caller is worthless. Shipping something that beats the Turing test for voice before the world has any way to prove who is real is reckless.
Pro-industry narrative
Voice is the most natural way to interact with machines, and a system that can listen while it talks finally feels like a conversation. Learning from speech alone, without a text crutch, captures the tone, pauses and timing that transcripts flatten. Running on a laptop keeps the audio at home instead of on someone else's server.
