Block releases Dessa, an AI model trained only on human speech
Jack Dorsey’s company Block has announced a new AI model named Dessa, which the team describes as a "true speech-to-speech" system. Unlike many current voice assistants that convert audio to text and back, Dessa is trained entirely on human conversation without any written text. The model is designed to capture the timing and emotional nuances of speech, aiming to address safety issues where…
Key points
- Block announced Dessa, a speech-to-speech AI model trained exclusively on human conversation.
- The model runs locally on a MacBook Pro M4 Max or better hardware.
- Block plans to give early access to a small group of AI safety researchers.
To demonstrate the model's capabilities, Block shared several chat logs and video clips. These examples show Dessa engaging in natural back-and-forth dialogue, though some interactions appear awkward or uncanny, such as initially mistaking a speaker for someone named Ronaldo. The team also highlighted a comparison where Dessa sounded more natural than OpenAI’s GPT-Live-1 voice model. Block states that the project was driven by curiosity, a desire for better human-AI understanding, and safety concerns regarding AI's inability to detect pain or emotion in voice.
The model is capable of running locally on a MacBook Pro M4 Max or better hardware. Block has stated that it will provide early access to a small group of AI safety researchers. The announcement emphasizes that speech is the most natural interface for intelligence, and that current voice AI often flattens this interaction by losing the subtleties of human timing and tone.
Model page: GPT-Live-1 →
Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend
The Verge AI · 23 October 2026
Loading the full article…
This text was published by The Verge AI and written by Jay Peters. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- Block Unveils Dessa, a Full-Duplex Speech Model Trained on ConversationPress · Unite.AI ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Moonshot AI releases Kimi K3.1 Pro with 1M-token context and open weights by Nov 4 · 3 src
- Ggml-org merges webgpu Q2_0 support into llama.cpp · 2 src
- Meta launches Muse frontier model family · 2 src
- OpenAI makes its Decisions API widely available using GPT-6 Luna · 6 src
- AgentionAI releases Qwen3.8-Flash-Next-Gyro-Coder for 24 GB GPUs · 1 src
Comments
via GitHub Discussions