Pinned Tweet
We reject: kings, presidents, and voting.
We believe in: rough consensus and running code.
—David Clark, 1992
We believe in: rough consensus and running code.
—David Clark, 1992
4,074
4,492
28,043
8,980,145
jack retweeted
the reasoning, latency, natural human-like tones, quality of audio, optionality, .. i am blown away. this is unlike anything i've seen on the audio model world.
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
6
7
49
32,239
jack retweeted
- Premium & cost efficient voice via Dessa
- Open source team-wide operations with Buzz
- Square UX for always-on agentic operations
- Both sides of the counter with Cash App Neighborhoods
Square about to usher the era of AI for neighborhood SMBs.
Let $XYZ cook
- Open source team-wide operations with Buzz
- Square UX for always-on agentic operations
- Both sides of the counter with Cash App Neighborhoods
Square about to usher the era of AI for neighborhood SMBs.
Let $XYZ cook
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
0
8
40
29,407
it’s getting great
Got setup w @buzzdotxyz beta at AGNTcon and super impressed with how modular & customizable it is w plugins.
Also very pleasant mellow design that sits pleasingly between corporate chaos of slack and wild west of discord. 🐝 💛
28
21
272
159,664
jack retweeted
Dessa is back
Even though this demo is impressive, I’ve been more impressed that Ragavan, @RayhaneMama, @AydinPolatDessa @marctyndel continue to show what a small dedicated team can do
We should raise our expectations of local models
Even though this demo is impressive, I’ve been more impressed that Ragavan, @RayhaneMama, @AydinPolatDessa @marctyndel continue to show what a small dedicated team can do
We should raise our expectations of local models
Meet Dessa, our AI model trained from scratch on nothing but speech.
We believe speech is the natural interface for technology. So we built Dessa.
And we’re inviting AI safety researchers to explore Dessa.
Learn more: block.xyz/dessa
10
8
76
31,636
jack retweeted
Lula vs. Bolsonaro: I was held in solitary confinement for 1,901 days for revealing the truth about the U.S. government's killings and espionage, including against Brazil and its industries. Out of power, Lula fought for my freedom. In power, Lula discussed my persecution with ed to meet with my family and offer help. After my release, Lula and I met in Rome to mourn the Holy Father's passing. Lula was kind to my children and asked for nothing in return. In contrast, Jair and Flávio Bolsonaro did nothing and said nothing. Yet without the ability to say no to Washington and Tel Aviv, there can be no domestic policy, only foreign instructions. Those orders are always the same: "Expand our theft of resources and wages, and keep your mouths shut about us murdering children." Sellouts will surrender everything in exchange for a cut, even their own tongues.
1,357
11,541
38,078
992,122
jack retweeted
Dessa no es otro modelo de voz. Es el primer intento serio de enseñar a hablar a una IA exactamente como aprendemos los humanos: solo escuchando.
Jack Dorsey acaba de presentar Dessa, el modelo de habla de Block entrenado únicamente con audio de conversaciones humanas. Sin texto. Sin modelos preentrenados. Sin convertir nada a letras. Audio entra, audio sale.
Eso cambia todo.
Hasta ahora casi todas las IAs de voz hacen lo mismo: escuchan → transcriben a texto → piensan en texto → sintetizan voz. En el camino pierden la risa a medias, el suspiro de frustración, el “ajá” que dice “te sigo”, la pausa que carga de emoción, el tono que delata ironía. Dessa se salta ese paso. Es full-duplex: escucha y habla al mismo tiempo. Puede interrumpirte, puede dejar que la interrumpas, puede hacer backchanneling (“sí, claro”, “¡no jodas!”) mientras tú sigues hablando. Habla con cadencia, emoción y silencios. Como una persona.
Y lo más loco: lo aprendió de la misma forma que un bebé. Solo escuchando. El equipo lo dice explícitamente: “Los bebés aprenden a hablar sin ver nunca una palabra escrita. Queríamos ver si un modelo también podía. Puede. Solo que es muy difícil”.
Además, es brutalmente eficiente. Se entrenó en un solo rack NVIDIA GB200 NVL72 y corre localmente en un MacBook Pro M4 Max o superior, más rápido que tiempo real. Un laboratorio pequeño dentro de Block, que reporta directamente a Dorsey, acaba de demostrar que no hace falta un cluster de 100.000 GPUs para avanzar la frontera del habla.
Todavía no está disponible para el público (solo acceso temprano para investigadores de seguridad y lista de espera), pero las demos que han soltado ya muestran algo que se siente distinto: más vivo, más torpe a veces, más humano.
Si esto escala, la forma en que hablamos con las máquinas va a cambiar de “interfaz de voz” a conversación real. Y eso, si lo piensas, es un cambio bastante profundo.
Jack Dorsey acaba de presentar Dessa, el modelo de habla de Block entrenado únicamente con audio de conversaciones humanas. Sin texto. Sin modelos preentrenados. Sin convertir nada a letras. Audio entra, audio sale.
Eso cambia todo.
Hasta ahora casi todas las IAs de voz hacen lo mismo: escuchan → transcriben a texto → piensan en texto → sintetizan voz. En el camino pierden la risa a medias, el suspiro de frustración, el “ajá” que dice “te sigo”, la pausa que carga de emoción, el tono que delata ironía. Dessa se salta ese paso. Es full-duplex: escucha y habla al mismo tiempo. Puede interrumpirte, puede dejar que la interrumpas, puede hacer backchanneling (“sí, claro”, “¡no jodas!”) mientras tú sigues hablando. Habla con cadencia, emoción y silencios. Como una persona.
Y lo más loco: lo aprendió de la misma forma que un bebé. Solo escuchando. El equipo lo dice explícitamente: “Los bebés aprenden a hablar sin ver nunca una palabra escrita. Queríamos ver si un modelo también podía. Puede. Solo que es muy difícil”.
Además, es brutalmente eficiente. Se entrenó en un solo rack NVIDIA GB200 NVL72 y corre localmente en un MacBook Pro M4 Max o superior, más rápido que tiempo real. Un laboratorio pequeño dentro de Block, que reporta directamente a Dorsey, acaba de demostrar que no hace falta un cluster de 100.000 GPUs para avanzar la frontera del habla.
Todavía no está disponible para el público (solo acceso temprano para investigadores de seguridad y lista de espera), pero las demos que han soltado ya muestran algo que se siente distinto: más vivo, más torpe a veces, más humano.
Si esto escala, la forma en que hablamos con las máquinas va a cambiar de “interfaz de voz” a conversación real. Y eso, si lo piensas, es un cambio bastante profundo.
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
17
48
395
67,706
jack retweeted
Back from AGNTCon + MCPCon with a better app out in the world 🍯
To kick things off @bradleyaxen provided the keynote. Then @kenlopez15 and @_tombrow demoed Buzz on mobile, so you can keep up with your agents wherever you go.
@WillPfleger and @tellaho showed how we build Buzz g, design, and agents shipping together in one shared space.
Finally @morganmartn, @wesbillman, @IsazaLucas, and the crew helped everyone get started building in the workshop. Thanks to everyone who came by, asked questions, and built with us.
And we announced Buzz 1.0. You can download it at buzz.xyz.
More coming soon 👀
To kick things off @bradleyaxen provided the keynote. Then @kenlopez15 and @_tombrow demoed Buzz on mobile, so you can keep up with your agents wherever you go.
@WillPfleger and @tellaho showed how we build Buzz g, design, and agents shipping together in one shared space.
Finally @morganmartn, @wesbillman, @IsazaLucas, and the crew helped everyone get started building in the workshop. Thanks to everyone who came by, asked questions, and built with us.
And we announced Buzz 1.0. You can download it at buzz.xyz.
More coming soon 👀
4
12
67
31,644
jack retweeted
probably one of the biggest releases this year tbh. absolutely insane! 🔥
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
2
6
64
33,353
jack retweeted
First AI anything that made me verbalize a “whoa” in quite awhile
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
3
4
53
41,148
RT @AiBreakfast: Ironically, imperfections in speech lead to perfectly natural results.
This is an order of magnitude better than anythin…
This is an order of magnitude better than anythin…
0
10
0
23,302
jack retweeted
I’m shook
OpenAI voice just got slapped hard
I’m in complete disbelief that this new voice model is made by Block and not Elevenlabs et al.
I mean Jack literally took it to 11 and solved the turing test for voice.
you’ll never know if you’re talking to a machine or human.
OpenAI voice just got slapped hard
I’m in complete disbelief that this new voice model is made by Block and not Elevenlabs et al.
I mean Jack literally took it to 11 and solved the turing test for voice.
you’ll never know if you’re talking to a machine or human.
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
114
127
1,668
238,771
jack retweeted
This is the most human voice AI I've heard.
Good job in capturing real human feel, even the laughs and random pauses we all do when we're thinking.
Block & Dessa team cooked. Well done!
Good job in capturing real human feel, even the laughs and random pauses we all do when we're thinking.
Block & Dessa team cooked. Well done!
59
53
569
72,241
team is amazing
What a legit team @blocks (@jack) put behind Dessa.
@RagavanThurair @RayhaneMama @carajzp @dvirzg.
• RealTalk (2019): the Joe Rogan voice clone
• Efficient-VDVAE
• Tacotron-2, 2.3K★
Now: full-duplex speech-to-speech trained only on raw speech.
61
24
537
237,239
jack retweeted
Let's do guessing game with our model!
Dessa developed world knowledge and learned to reason purely from audio
Here's our intern @carajzp playing a game with it
4
10
56
30,076
jack retweeted
Damn this is stunningly good.
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
13
14
293
90,141
exactly
This is the future of how small businesses will be able to talk to their customers and interact with their own tools.
Block is building the entire stack for commerce on Main Street, from AI kiosks and agentic payment rails to voice models that run on edge devices. Very cool.
59
43
1,139
295,992