Early Beijing time on October 24, 2026 (UTC evening October 23), Block put Dessa on its official page and posted from @blocks asking AI safety researchers to get involved. Dessa takes audio in and returns audio out. Full duplex means it listens and talks at the same time. Block says training used only human conversation audio—no text corpus, no pretrained base model. On the page, Block hedges that, as far as it knows, Dessa is the world’s first AI you can have a realistic live conversation with that was trained only from human speech.
It's still in a trial stage and hasn't been released to the public yet. Only a small group of AI safety researchers get early access; everyone else needs to apply on the official site. Block also says inference runs faster than real time, that it can be deployed locally on a MacBook Pro M4 Max or better, and that training used one NVIDIA GB200 NVL72.
The team is a tiny AI research lab inside Block, started by former members of Dessa, the Toronto deep-learning startup, and reports straight to Jack Dorsey. As for whether it performs locally the way the official demos claim, or how safety researchers have been trying it so far—we don't have that information yet. If you've used it, you can send feedback through the site-owner contact link in the footer.