Block’s AI Lab Releases Dessa, a Voice-Only Conversational Model That Runs Locally

Dessa's significance lies in its from-scratch, speech-only training and local execution, which sidesteps text-based pipelines and cloud dependence—a departure from mainstream voice models. The limited release to safety researchers signals caution about misuse, while the hardware requirement restricts practical access. Whether this approach scales beyond a research preview and how safety findings shape broader availability are the open questions worth watching.