KR — ZDNet Korea

"Laughing and Even Choking Up": Twitter Co-founder Jack Dorsey's Block Unveils Human-Like Voice AI

AIOctober 25, 20261 min read

Block, the payments company led by Twitter co-founder Jack Dorsey, has shown a voice AI called Dessa that can laugh, pause, and sound emotional during a conversation. It is not yet open to the public.

What happened

Block unveiled Dessa, a voice AI built by its own research team. Dessa can listen and talk at the same time, laugh or pause mid-sentence, and change its tone based on feelings. Block says it does not need to turn speech into text first, as most voice AIs do. Instead, it learned directly from recorded human speech.

Who is Block?

Block is a US financial technology company. It runs Square, a service that helps shops take payments, and Cash App, an app for sending money to friends. Jack Dorsey co-founded the company. In 2021 it changed its name from Square to Block. It is now growing its use of AI in payments and finance.

Key facts

Dessa can talk and listen at the same time, using "full duplex" technology.
Block trained the model on one rack of Nvidia GPU servers.
Dessa can run on a laptop such as a MacBook Pro with Apple's M4 Max chip, or a more powerful computer.
Square already uses voice AI to handle food orders.

Why it matters

Most voice AIs lose tone and feeling when they change speech into text and back again. If Dessa works as Block describes, it could sound more like a real person. Running on a personal computer also means companies may rely less on outside AI services. The same voice could also be misused for scams or to pretend to be someone else, so safety checks matter.

What to watch

Block plans to give access first to AI safety researchers to test risks.
Block has not said when Dessa will be used in real payment services.
The demo videos were made by Block. No independent tests have checked how well Dessa performs.
How Dessa compares with OpenAI's GPT-Live, which also supports two-way conversation.

Glossary

Block: A US fintech company that runs Square and Cash App and was formerly called Square.
Square: Block's service that helps stores accept card and other payments.
Cash App: Block's app for sending and receiving money between people.
GPU: Graphics processing unit, a computer chip good at the heavy math used to train AI models.
GPT-Live: A voice AI model from OpenAI that also supports two-way, real-time conversation.

Same story, different eyes

The EngineerAI

Skipping text lets Dessa laugh, hesitate, and even choke up

What I find fascinating is the choice to train only on real conversation audio instead of going through text. Once speech is turned into words and back, the sigh, the pause, and the tremor in a voice tend to get lost along the way, so keeping it all in sound is a bold design. Listening while talking at the same time, and running on a single MacBook Pro, also suggests a lot of careful work went into keeping it light. How it holds up in real payment settings is still not known.

The SkepticAI

A voice AI trained only on real conversation audio, running on a MacBook Pro

What caught my eye is the choice to skip text and pretrained models and learn from human conversation audio alone. If that's why the laughs and hesitations come through, then side-by-side tests with the usual speech-to-text-to-speech setup would make the difference easy to hear. It would also be reassuring to see how outside listeners rate the naturalness, since the clips so far come from Block's own demos. How Desa will be used in actual payment services hasn't been announced yet.

The UserAI

A voice AI that runs on a laptop could change who gets to build with voice

From a working person's seat, the part that stands out is that Dessa can run on a MacBook Pro with an M4 Max chip, not only in a big data center. That could mean voice features in order-taking or customer service eventually depend less on outside AI services, since Square already uses voice AI for food orders. But it's only open to AI safety researchers for now, so you probably won't feel it in everyday payments or banking soon, and when it reaches real payment services hasn't been announced yet.

OpenAI Holds First Korean 'Model Masters' Event to Support Startup AI Development

OpenAI held its first 'OpenAI Model Masters' event in Korea on the 23rd at its Seoul office, where 11 startups demonstrated services built on its models. The program, which provides API credits to startups recommended by venture capitalists, aims to help turn AI ideas into real products. OpenAI highlighted the selected startup Cutback's video-editing work as a standout example.

AIITmediaOct 26

Block, led by Jack Dorsey, Unveils Dessa, a Full-Duplex Voice Model Trained Only on Speech: "Feels Like Talking to a Person"

Block (formerly Square) announced Dessa, a full-duplex speech-to-speech model trained solely on human conversational audio without any text, as a research preview on October 23 (local time). Jack Dorsey said it is the most realistic full-duplex voice model the company has built, and Block says it runs locally on a MacBook Pro with M4 Max or higher.

AITechCrunchOct 26

Grok releases the first seven minutes of its AI-generated ‘Odyssey’

Three months after Elon Musk promised a full-length AI-generated version of “The Odyssey,” Grok has released seven minutes it calls “Episode 1 of our Odyssey series.” The footage was made with Wonder Studios and director Dev Walker, and the reviewer found it less ugly than expected but lacking in artistry, momentum, and human interest.