🔍 Read the full analysis: AI Voices Now Laugh And Interrupt. Why That’s Not Proof Of Anything on ThorstenMeyerAI.com
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Block introduced Dessa, a speech-to-speech research model whose company-selected demos show laughter, emotional cues and conversational interruptions. As of Oct. 26, outsiders had no public access, technical paper or benchmarks, so the recordings do not establish how reliably it works or whether it understands emotion. Its humanlike style also sharpens concerns about voice impersonation, while existing fraud already uses cloned voices.
Block, the company behind Square and Cash App, announced Dessa on Oct. 23. The company says the model was trained from scratch on recorded human conversation, without text or pretrained models. It describes Dessa as full-duplex: the system can listen while speaking, respond to pauses and cut in or backchannel rather than waiting for a turn to finish.
Block’s project page features recordings in which the model laughs mid-sentence, responds to a surprise and helps a speaker recall a film title. The company says the clips are live conversations that were not steered. Those recordings show what Block chose to publish, but cannot establish performance across other speakers, topics or conditions.
Block also says inference runs faster than real time on a MacBook Pro M4 Max or better, and that training used one NVIDIA GB200 NVL72. It has not supplied a latency measurement or disclosed the model’s size. The model is not publicly available; Block says a small group of AI safety researchers will receive early access, but has announced no weights or API.
Infographic 01 · Dessa fact sheet
What Block claims, what was heard, what is missing.
Every capability below comes from Block or its selected recordings. Nobody outside Block has run the model yet.
| Topic | What we know | Status |
|---|---|---|
| Training data | Human conversation only: “no text, no pretrained models” | Block’s claim |
| Conversation | Full-duplex: listens while it talks, backchannels, cuts in | Block’s claim |
| Emotion | Hears and speaks laughter, sarcasm, sighs; unclear if steered by tags (TechCrunch) | Block’s claim |
| Speed | “Faster than real time”; no latency figure | Block’s claim |
| Hardware | Trained on one NVIDIA GB200 NVL72; runs on a MacBook Pro M4 Max or better | Block’s claim |
| Demos | Some chats “a little awkward and uncanny”, more natural than GPT-Live-1 (The Verge) | reporters heard it |
| Size, benchmarks, release date | Not disclosed (CoinDesk) | not disclosed |
| Paper, model card, license, weights | None found as of 26 Oct 2026 | not disclosed |
| Languages | Not listed; Block calls learning new languages an open question | not disclosed |
| Voice cloning safeguards | No statement on watermarking, consent checks or imitating a specific person | not disclosed |
Who can use it
research preview
- Public: a waitlist; “launch links are coming soon”
- Early access: “a small group of AI safety researchers”, institutional email required
- Business use: none announced; Square already uses voice AI for ordering (TechCrunch)
Who built it
from Block’s page
- A “(very) small AI research lab at Block” reporting to Jack Dorsey
- Founded by former members of Dessa, the Toronto startup Square acquired in 2020 (CoinDesk)
- Lead Ragavan Thurairatnam: “a GPT3 moment for speech”
Sources: Block, Dessa project page (23 Oct 2026); Ragavan Thurairatnam on X (23 Oct 2026); The Verge (23 Oct 2026); TechCrunch (24 Oct 2026); CoinDesk (24 Oct 2026).
Powered by Thorsten Meyer AI
Why Natural Voice Raises Stakes
More lifelike speech can make voice interfaces easier and more comfortable to use. It can also make it harder for a listener to tell whether a caller is human or synthetic. That concern is not proof that Dessa can impersonate a person: Block says it chooses its own voice and personality, and the company has not said whether it can reproduce a specific person’s voice.
The risk exists beyond this model. The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI, with $893 million in reported losses. It also reported more than $5 million in losses from distress scams using cloned voices to imitate loved ones. These figures describe reported complaints and losses across AI-related crimes; they do not show that Dessa caused any fraud.
For businesses and families, the practical point is that a familiar voice alone is weak verification when money, account access or urgent requests are involved. Independent confirmation through a known phone number or another trusted channel offers a stronger check than judging whether a voice sounds natural.
Products Worth Considering
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
Anker SOLIX C200 DC Portable Power Station, 192Wh, 200W
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
What Block Has Demonstrated
Block’s announcement draws on the work of a small research team associated with Dessa, the Toronto deep-learning startup acquired by Square in 2020, according to CoinDesk. Block says the model learns directly from speech rather than from a text-based language model converted into voice. That design is intended to preserve features such as laughter, sarcasm and pauses.
Coverage of the recordings has been more cautious than the launch language. The Verge described some exchanges as “a little awkward and uncanny,” while saying Dessa came across as more natural than another model in the comparison. TechCrunch said it was unclear whether emotional qualities might be steered by tags. CoinDesk reported that model size, benchmark results and a release date had not been disclosed. These accounts assess the published clips, not an independent test of the model.
The distinction matters: a system may produce convincing emotional cues without experiencing emotion or reliably interpreting a person’s state. Block argues that speech systems often miss emotion, which can make interactions less responsive. Whether Dessa can detect emotional meaning consistently, across languages and real-world calls, has not been established.
““Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.””
— Jack Dorsey, Block co-founder, as quoted in Block’s announcement
What Independent Tests Must Establish
Key technical questions remain open. Block has not published a technical paper, model card, benchmark results, license or release date. The company has not provided a list of supported languages or a measured latency figure, and it has identified learning new languages as an open question.
It is also unclear how the model handles consent, whether its audio is watermarked or otherwise identifiable as synthetic, and whether it can imitate a particular person. No outside party has yet tested whether the demos generalize to unfamiliar voices, interruptions, noisy calls or sensitive situations. The clips do not prove that Dessa has humanlike understanding, nor do they establish that it is safe or unsafe for a particular use.
Release and Safety Checks Ahead
The next meaningful step is access for the small group of AI safety researchers Block says will receive early versions. Their testing could help clarify performance, limitations and safeguards, but the company has not published a timetable or a public-access plan. Any technical claims should be judged against reproducible results once those are available.
Until then, organizations handling payments or sensitive requests can use callbacks to previously verified numbers and require separate confirmation for urgent transfers. Families can agree on a code word and end a suspicious call before calling back using a saved number. These precautions address voice-cloning scams generally; they are not evidence that Dessa has been used in one.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says can listen and talk at the same time. Its published demos include laughter, emotional reactions and interruptions.
Has anyone independently tested the model?
As of Oct. 26, 2026, no outsider had access to test it. The public evidence consists of recordings selected and published by Block.
Does expressive speech show that an AI feels emotion?
No. A voice can produce emotional-sounding cues without evidence that the system experiences emotion or understands a person’s feelings.
Can Dessa clone a specific person’s voice?
That has not been established. Block says Dessa chooses its own voice and personality, but has not publicly explained whether it can imitate a specific person.
How can people check an urgent call asking for money?
Hang up and call the person back using a number already saved or independently verified. Families can also agree on a private code word for emergencies.
Source: ThorstenMeyerAI.com
