Block introduced Dessa, which it says was trained from scratch on human conversations without text or a pretrained language model. Direct audio processing and simultaneous listening and speaking aim to handle intonation, laughter and interruptions. The company says it runs locally on an M4 Max MacBook Pro, but access is currently limited to selected safety researchers. Curated recordings and company statements do not establish broad performance or genuine emotional understanding, and the model should not be mistaken for a public release.