Block unveils Dessa, a full-duplex speech model trained only on human conversation
daluoseo · x · 2026-10-26
Block.xyz has launched Dessa, claiming it's the first AI trained from scratch purely on human speech — no text, no pretrained models — enabling realistic real-time full-duplex conversation. Demo recordings show it laughing mid-sentence at jokes (rather than on cue), speaking louder when asked, completing a person's sentence about a movie title, and even producing synchronized dual-speaker tracks while choosing its own voice and personality. The team argues voice AI that converts speech to text loses emotion and nuance, and positions Dessa as a new research frontier: how efficiently can speech-only models learn language, facts, new languages, and in-context skills? Waitlist is open.
More from Models
- DeepSeek V4.1 Flash with OpenCode Go: cheap unlimited coding agent with Opus validating results — haider1 · 2026-10-26
- yacine claims OpenAI models have allowed 'borderline and dangerous' tasks since codex 5.4 — yacineMTB · 2026-10-26
- DeepSeek V4.1F tops Chinese models on FrontierMath Tier 4 at 46% — teortaxesTex · 2026-10-26
- Casado slams Claude's safety classifier for flagging 'fetching my own documents' as reasoning extraction — AccBalanced · 2026-10-26
- Yacine MTB: DeepSeek is 'undoubtedly' a frontier model — yacineMTB · 2026-10-26
- Single-neuron steering: injecting a "joyful neuron" visibly alters a model's self-description — Sauers_ · 2026-10-26