Block's new speech-to-speech AI Dessa, trained purely on conversation, chats awkwardly
The Verge AI · rss · 2026-10-24
Block, Jack Dorsey's company, announced Dessa, billed as the first AI model trained entirely from scratch on human speech for realistic full-duplex voice conversation.
To showcase it, Block shared real chats between Dessa and humans — sometimes awkward and uncanny. In one conversation shared by Dessa lead Ragavan Thurairatnam, the model mostly held a natural back-and-forth, though it initially mistook the speaker for someone named "Ronaldo".
More from Multimodal
- ElorianAI's Glimpse multimodal model IDs Ed Sheeran concert from blurry video, claims to beat ChatGPT at fraction of cost — AndrewDai · 2026-10-24
- Letting Claude run 10+ hours to auto-build full ComfyUI workflows — ThatKnarfGuy · 2026-10-24
- Ideogram Vector + Opus 5.5: one prompt to editable SVG, then a full animated explainer — CSProfKGD · 2026-10-24
- Striking MidJourney --sref demo turns Superman into motion-blurred abstract art — gizakdag · 2026-10-24
- Early Kimi K3.1 Pro access: $4 in tokens, 3-4 hours for Blender clay renders — ChrisGPT · 2026-10-24
- MiniMax H3 Chain Director V1.0 chains multi-segment video generation in ComfyUI — Far_Patience4517 · 2026-10-24