Block unveils Dessa, a text-free full-duplex speech model that runs locally on M4 Max
lmoroney · x · 2026-10-24
Block introduced Dessa, a speech-to-speech model trained from scratch on human speech alone — no text data, no pretrained models. It is full duplex (listens and talks simultaneously), reportedly runs locally on a MacBook Pro with an M4 Max, and Block is inviting AI safety researchers to explore it. It's also a great case study in how training data shapes capability: pauses, laughter and interruptions all live in the audio itself.
More from Models
- Developer burns 12M DeepSeek tokens for just 16 rupees, marvels at the price — cneuralnetwork · 2026-10-24
- SuperGrok Ultra costs $200/mo on the web, $260 via app stores due to 30% platform tax — elonmusk · 2026-10-24
- Do CLIP models get updated over time? Reddit user asks about drifting prompt behavior — apostrophefee · 2026-10-24
- "Guardrails make agents stupid": Dev renews push for uncensored LLMs — AIFlow_ML · 2026-10-24
- Musk spotlights Grok Bot Ultra at $259.99/month: 'crazy until you see the time it saves' — elonmusk · 2026-10-24
- Same model, 2.3x cost gap: cache hit rates make per-token pricing a poor guide — nicolechirps · 2026-10-24