Dessa AI Twitter & X Wall: Official Announcements and AI Community Buzz

jack

@jack · Block Head & Cofounder, Block, Inc.

X Official Launch

“We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence. block.xyz/dessa”

Block cofounder Jack Dorsey unveils Dessa AI, framing pure speech-to-speech interaction—with emotion, pauses, and natural cadence—as the true interface for intelligence.

Ragavan Thurairatnam

@RagavanThurair · Dessa AI Lead at Block (Former Cofounder, Dessa Toronto)

X Lead Researcher Thread

“We made the first AI model you can have a realistic conversation with that was trained entirely on speech! Dessa is a full-duplex speech-to-speech model that you can run on consumer GPUs. This is a GPT-3 moment for speech.”

Dessa project lead Ragavan Thurairatnam calls Dessa a 'GPT-3 moment for speech'—the first realistic live conversational model trained 100% on raw speech with zero text.

Ragavan Thurairatnam

@RagavanThurair · Dessa AI Lead at Block

X Live Audio Demo

“What happens when you put Dessa on the phone with GPT-Live-1? Nobody steers it. Listen to how Dessa drops natural backchannels ('okay', 'right', 'that's relieving') while GPT-Live delivers textbook paragraphs.”

Highlighting the direct unsteered conversation between Dessa AI and GPT-Live-1: Dessa's organic sub-second backchanneling contrasts sharply with text-pipeline voice assistants.

AI Systems & Compute Thread

@RagavanThurair · Block Deep Learning Lab Engineering Note

X & Hacker News Top Thread

“Trained from scratch on a single NVIDIA GB200 NVL72 rack. Runs locally faster than real-time on a MacBook Pro M4 Max or better. Zero text tokens. Hundreds of times fewer words than text LLMs.”

Engineers on X and Hacker News zeroed in on two stunning numbers: Dessa was trained on just one NVIDIA GB200 NVL72 system and runs faster-than-real-time locally on an Apple M4 Max.

Speech ML Research Commentary

@RagavanThurair · Acoustic Representation & Prosody Breakdown

X Viral Quote Breakdown

“Say 'I didn't say she stole my CUDA code' out loud, stressing a different word each time. Same 7 words, 7 completely different meanings. Run speech through STT -> Text LLM -> TTS and all of that acoustic truth is destroyed.”

The 'CUDA code' sentence stress test became the most quoted passage from the Dessa launch paper—illustrating why intermediate text bottlenecks strip away human intent.

AI Alignment & Voice Safety Forum

@RagavanThurair · Early Access Call for AI Safety Researchers

X Safety Research Call

“AI keeps gaining power and independence, and it's still nearly deaf to the emotion in our voices. Imagine being in serious pain while a voice AI has no idea and just keeps going. That's the safety problem. Calling all safety researchers!”

Why Block restricted Dessa AI to a safety researcher preview: a powerful AI that cannot hear distress, pain, or trembling in a human voice poses a fundamental alignment hazard.

Voice UX & Duplexity Breakdown

@RagavanThurair · Unsteered Audio Sample Reaction

X Demo Clip Spotlight

“It laughs the moment something's funny, mid-sentence, the way people do. Other models wait for you to finish, then laugh on cue. Ask it to speak up and it gets louder. Nobody taught it that.”

Community reaction to the 'Talking about the release' clip: emergent vocal amplitude control ('speak up!') and spontaneous mid-sentence laughter without prompt engineering.

Multimodal Benchmark Watcher

@RagavanThurair · AI-to-AI Collision & Turn Recovery Analysis

X Audio Benchmark Thread

“In the Noisy GPT Live sample at 4:07, both AIs speak at the same time. GPT-Live-1 triggers a hard stop ('Sorry, go on.'), while Dessa smoothly replies: 'Oh, no, that's perfectly fine. I like what you're saying...'”

A side-by-side look at 4:07 in the Noisy GPT-Live recording: how full-duplex audio models negotiate speech collisions gracefully versus hard VAD interrupts.