AI Can Laugh And Interrupt Like A Person. Block’s Dessa Shows How

🔍 Read the full analysis: AI Can Laugh And Interrupt Like A Person. Block’s Dessa Shows How on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Block has demonstrated Dessa, a research model that the company says can listen and speak at the same time, laugh during a conversation and respond to interruptions. The October 23 recordings offer a glimpse of more natural-sounding voice AI, but the model remains unavailable for independent testing, and Block has not released technical documentation or a public access date.

Block says Dessa was trained from scratch on recorded human conversation, without text or pretrained models. The company says the system handles speech in a full-duplex mode: it can listen while speaking, decide when to cut in, and provide conversational responses such as backchanneling. Block also says it can recognize and produce cues including laughter, sarcasm and a frustrated sigh.

The demonstrations on Block’s project page include a model laughing mid-sentence, helping a speaker remember a film title and responding to an anniversary surprise. Block describes the clips as live conversations that were not steered. They are company-selected recordings, however; no outside researcher or journalist has publicly run the model. The Verge reported that some exchanges felt “a little awkward and uncanny,” while finding Dessa more natural-sounding than another model in a comparison.

Block says inference runs faster than real time on a MacBook Pro M4 Max or better, but did not provide a latency figure. The company also says it trained the model using one NVIDIA GB200 NVL72 and substantially fewer words than large language models typically require. Those are company claims: as of October 26, there was no published paper, model card, benchmark, license, released model weights or API to verify them.

More Natural Voices, New Fraud Risks

Dessa matters because conversational timing and vocal expression are part of what people use to judge whether they are speaking with a person. A voice that pauses, laughs or reacts naturally may make customer-service systems easier to use. The same qualities could make a synthetic voice harder to distinguish from a real person during a call, particularly when the caller creates urgency or claims to be a relative or colleague.

The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI and $893 million in reported losses. The report also recorded more than $5 million in losses from distress scams using cloned voices to imitate loved ones. These figures describe reported complaints and losses, not the frequency or effectiveness of Dessa-like systems; there is no evidence in the source material that Dessa has been used in fraud.

The practical concern is not that this model has been shown to enable a new crime, but that voice alone is a weak identity check when synthetic speech can sound more natural. For businesses, that points to verifying payment requests through a separate, trusted channel. Families can agree on a code word and call back using a number they already have rather than relying on a caller’s voice or incoming number.

How Block Presented the Research

Block, the company behind Square and Cash App, published Dessa’s project page on October 23, 2026. The work comes from a small Block research lab founded by former members of Dessa, a Toronto deep-learning startup acquired by Square in 2020, according to CoinDesk. Block says the model learns directly from conversation rather than relying on a text-based language model and a separate speech-generation pipeline.

Block co-founder Jack Dorsey described talking with the system as “like talking with a person,” citing emotion, nuance and natural cadence. Dessa lead Ragavan Thurairatnam called the work “a GPT3 moment for speech,” and said it could run on consumer GPUs. Those comments express the team’s view of its research; they are not independent evaluations of performance.

Access remains limited. Block says a small group of AI safety researchers has early access, while other interested users can join a waitlist. TechCrunch reported that it was unclear whether some emotional effects in the recordings were directed by tags. CoinDesk said Block had not disclosed the model’s size, benchmark results or a public release date. The Decoder cautioned that sounding emotional does not establish that a system is emotional or intelligent.

“Talking with it feels like talking with a person… full of emotion, nuance, pauses, and a natural cadence.”

— Jack Dorsey, Block co-founder

Independent Tests and Safeguards

Key technical and safety details remain undisclosed. There is no public paper or model card describing the training data, evaluation methods, limitations or error rates, and no independent test of Block’s claims about conversational ability, speed or hardware requirements. Block has not said when the model might be publicly released or provided a list of supported languages; the company has described learning new languages as an open question.

It is also unclear whether the model can imitate a specific person, what consent checks might apply, or whether outputs will carry a watermark or other disclosure. Block says Dessa selects its own voice and personality, but that statement does not answer whether safeguards prevent targeted imitation. The demonstrations do not establish that the model has been used for scams or that its synthetic speech is indistinguishable from a person’s voice.

Release Details Still Pending

The next concrete developments would be a technical paper, a model card, independent evaluations and details of any wider release. Those materials could clarify the model’s performance, hardware demands, language coverage and protections against voice imitation. For now, Block’s access is limited to a small group of safety researchers and a waitlist, with no public release date announced.

Until more evidence is available, the recordings should be treated as demonstrations of what Block says its system can do, not a validated measure of real-world reliability or fraud risk. Organizations handling payments or sensitive requests can review their existing call-verification procedures without assuming that Dessa itself is in use.

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, respond to conversational cues and produce expressive speech. It is not publicly available for general testing as of October 26, 2026.

Has anyone independently verified Block’s claims?

Not publicly. The available evidence consists of recordings selected by Block. No public paper, model card, benchmark or outside test was available by October 26.

Can Dessa imitate a specific person’s voice?

That remains unclear. Block says the system chooses its own voice and personality, but has not publicly detailed voice-imitation limits or consent safeguards.

Does the announcement show that Dessa has been used in scams?

No. The source material reports no such use. The fraud concern is a potential risk of more natural synthetic speech, discussed alongside FBI figures on reported AI-related complaints and voice-cloning distress scams.

How can people verify an urgent voice call?

For requests involving money or sensitive information, use a separate trusted channel: hang up and call a known number. Families can also agree on a private code word rather than treating a familiar-sounding voice as proof of identity.

Source: ThorstenMeyerAI.com