News for hackers. Stuff that matters.

A comment on Dessa: the most authentic full-duplex speech model ever built.

Re:Dessa: the most authentic full-duplex speech model ever b... by moonu on Friday October 23, 2026 @04:50PM (#50215143)
Can you instruction-train a duplex model like this? How would that look? I also wonder, if a model trained only on speech can develop world knowledge like this, could you train an image model only on images of text to essentially act as an LLM?
[ Parent | on HN ]
Re:Dessa: the most authentic full-duplex speech model ever b... by k141 on Sunday October 25, 2026 @12:54PM (#50233587)
yeah, there are models for that already, some are interesting, some are not. check out PEXEL and DeepSeek-OCR.

in theory this could cut costs pretty well for long contexts + there no tokenisation layer.

[ Parent | on HN ]