Block Releases the research preview of the Dessa voice dialogue model, which can run locally on a MacBook
BIBIBI
AT A GLANCE
Block Released the Dessa voice model, saying it can run locally in real time on a high-specification MacBook.
Article
Block Releases Dessa, a dialogue model trained solely on speech, which can run locally on a MacBook
PANews October 24 reported that, according to Block official sources, the fintech company led by Jack Dorsey Block’s AI research lab released the research preview of Dessa, a full-duplex voice dialogue model. The model was trained from scratch entirely on human conversational speech, without using text or pretrained models, and can listen and speak simultaneously; training used only one set of NVIDIA GB200 NVL72, and it can run locally at faster-than-real-time speeds on a MacBook Pro M4 Max or higher-specification device. Dessa is not yet broadly available. The company has released recordings of dialogue demonstrations and provided early access to a small number of AI security researchers. Project lead Ragavan Thurairatnam called it “the GPT-3 moment in voice.”
Block’s AI research lab released the research preview of the full-duplex voice dialogue model Dessa.
02
Dessa was trained from scratch entirely on human conversational speech, without using text or pretrained models.
03
Dessa can listen and speak simultaneously.
04
Training used only one set of NVIDIA GB200 NVL72。
05
Dessa can run locally on a MacBook Pro M4 Max or higher-specification device at faster-than-real-time speeds.
06
Dessa is not yet broadly available and has only been provided to a small number of AI security researchers for early access.
07
Project lead Ragavan Thurairatnam called it “the GPT-3 moment in voice.”
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
Dessa is a dialogue model trained solely on human speech that can listen and speak simultaneously. It reportedly can run locally on a high-specification MacBook, but it remains a research preview and is not yet fully available to general users.
Why it matters to readers
This shows that Block is researching a voice dialogue model that runs locally; however, publicly available information currently focuses primarily on the research preview and demonstrations and has not yet indicated productization or commercial applications.
Local operation means the model can process conversations directly on a personal computer and may not need to connect to cloud servers; “full-duplex” means it can listen and speak simultaneously.
Risks and unknowns
The evidence comes from PANews’ account of Block official sources and does not provide Block a link to the original announcement.
Dessa is not yet broadly available, and the scope of public testing is limited.
The model’s performance, safety, and demonstration results still require validation through further independent evaluations.
The evidence does not specify productization plans, commercial applications, or revenue.
Related Developments
Loading event timeline…
Related concepts
full-duplex
This term is not in the glossary yet. Browse related concepts in the glossary.
