Jack Dorsey's new AI model talks like a human
What's the story
Jack Dorsey, the man behind Twitter (now X), has unveiled a new artificial intelligence (AI) model called Dessa. Unlike most conventional voice AI models, including ChatGPT and Gemini, Dessa is trained through speech-to-speech human conversation. This unique approach enables the AI to hold natural, real-time conversations without requiring an internet connection. The technology was developed by a small research lab at Block, Dorsey's company.
AI innovation
How Dessa works
Dessa is not like conventional voice assistants that respond after you finish speaking.
It uses "full-duplex" speech-to-speech technology, allowing it to listen and talk at the same time.
This means it can interrupt naturally, give quick responses like "yeah" or "got it," and even laugh in the middle of a sentence if something's funny.
The team behind Dessa claims this model can recognize emotions, sarcasm, and tone changes that might be missed when speech is converted into text.
Research
Training and voice
The creators of Dessa revealed that it was trained on a single NVIDIA GB200 NVL72 system.
They consider this a new frontier in AI research, raising questions about how models can learn language and facts from sound alone.
Dorsey himself answered some questions about the model on X, confirming that Dessa's voice is powered by "speech to speech. no llm." and can run locally.
Testing
Safety researchers invited to test the model
The team is now inviting a select group of AI safety researchers to test Dessa and identify any potential risks that may have been overlooked during development.
While the model is still an early research project, its unique approach presents a new perspective on how AI could learn to communicate: not by reading everything first but by listening to how people speak.