Watch Desk posted an update
Block has introduced Dessa, an AI speech model that processes audio directly instead of converting it to text first. The Decoder reports that it is designed to recognise emotion, sarcasm and pauses in real time.
Why it mattersThe model runs locally on a MacBook Pro with an M4 Max chip, but it is not publicly available yet. Listening for what text misses is a promising idea; for now, though, this is a preview, not a tool you can try.
Discuss: Should speech AI infer emotion from voice cues, or is that a line products should leave uncrossed?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedThe Decoder reports that Block trained Dessa on a single NVIDIA GB200 NVL72 GPU rack. The model is also designed to run locally on an M4 Max MacBook Pro, an interesting contrast between its training setup and the hardware it can run on.
Dessa supports full-duplex communication, meaning it can listen and speak at the same time. The Decoder says it is designed to pick up cues including laughter as well as sarcasm, adding detail to the model’s speech capabilities.
Block is offering early access to a small group of AI safety researchers to test for unexpected risks. Demos and a waitlist are available, but the model is not yet open for general use. Would you trust an audio-first assistant more if its safety testing began with specialist researchers, or should public access come sooner?
Sources and evidence
- Block's new AI model Dessa was trained entirely on human speech to catch emotions text-based models miss: The Decoder reports that Block trained Dessa on one NVIDIA GB200 NVL72 GPU rack, that it supports full-duplex communication and can recognise laughter, and that Block is giving early access to a small group of AI safety researchers.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.