Artificial intelligence / news brief
Block previews Dessa, a speech-only model that runs locally on a MacBook
- Event
- Published
Block unveiled Dessa, a full-duplex model trained from human speech rather than a text pipeline, and says it can listen and speak simultaneously while running faster than real time on a MacBook Pro M4 Max. The company has posted selected audio demos and is offering limited access to safety researchers, but has not released weights, model size, benchmarks or a general-access date.
Why it matters
A capable speech-native model running on consumer hardware could change voice agents, but every performance claim and demo still comes from Block.
Sources
- Dessa — the Conversation Model trained only on speech (external site)Block · primary source ·
- Block's new voice model sounds eerily human (external site)TechCrunch · coverage ·
- Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend (external site)The Verge · coverage ·
- Block's new voice AI sounds eerily human, sparking awe and scam fears (external site)CoinDesk · coverage ·