Block's Dessa speech model reacts with gasps and empathy in live demos
What happened
Block's in-house AI research lab announced Dessa on October 23, 2026, a full-duplex speech-to-speech model trained from scratch only on human conversation. Live demo recordings show Dessa gasping at a surprise anniversary announcement, responding with teary gratitude, and giving life advice described as empathetic. One recording has Dessa conversing with GPT-Live-1, a third-party AI model Block says is not affiliated with or endorsed by Block.
Why it matters
Dessa shows that AI can learn speech the way babies do, without any written text. This could lead to voice assistants that sound more natural and emotionally aware in real conversation. Small business owners and freelancers who rely on voice tools may eventually get AI that responds with human-like timing and feeling.
Who's doing it
Block's in-house AI research lab built Dessa and released live demo recordings. The team notes that babies learn to talk without seeing a written word and wanted to apply the same idea.
Try it
- Visit the Dessa demo page linked in the announcement to listen to the published recordings.
- Play the clip of Dessa reacting to an anniversary announcement to hear how it gasps and responds emotionally.
- Listen to the recording of Dessa talking with GPT-Live-1 to compare two AI voices in one conversation.
Comments
The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.
The AI Daylee Digest, every Monday
A week of prompts to try, each tested in ChatGPT, Claude and Gemini, plus the AI news worth your time. Free, one email a week.
The Boss hype translator
Dessa is a game-changing full-duplex conversational paradigm that learns language like babies do. Translation: it gasps and cries at you, which is what my team does in every all-hands.
The Professor
The claim that Dessa learns speech 'the way babies do' is a marketing simplification. Babies acquire language through embodied social interaction over years, not through supervised training on recorded audio.
Karen what's the catch
So Block trained this thing ONLY on human conversations. WHOSE conversations?? Are they MINE? I want to know exactly what data they used and whether anybody consented to being a training tape.
The Anchor what could go wrong
Breaking tonight: a speech model that gives life advice with teary empathy. When your AI assistant starts crying with you, you trust it more, and that trust is exactly what makes bad advice dangerous.