Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend

The AI model Dessa, created by Block's Jack Dorsey, speaks with humans in a manner reminiscent of an awkward new acquaintance. The model, trained "from scratch on nothing but human conversation," has been showcased by Block through various conversations with individuals, some of which have been described as awkward and uncanny.

For instance, one conversation was shared by Dessa lead Ragavan Thurairatnam, which involved the model and speaker engaging in a mostly natural dialogue, except for an initial mistake where the model misidentified the speaker as someone named "Ronaldo." The Dessa website provides several examples of the model's interactions, including a guessing game played via voice, a discussion about Christopher Nolan's films, a person's account of taking Dessa on a surprise trip to Paris, and a conversation between OpenAI's GPT-Live-1 voice model and Dessa, where the latter comes across as more natural.

According to the Dessa team, the AI model is a "true speech-to-speech model," meaning everything it knows has come from human speech. It can run locally on a MacBook Pro M4 Max or better. The team behind Dessa provided three reasons for its creation: curiosity, better understanding between humans and AI, and safety. The learning process behind Dessa is said to be similar to how babies learn to talk without ever seeing a word written down.

The team behind Dessa believes that speech is the most natural interface to any intelligence, super or otherwise, and that today's voice AI often wastes this potential by flattening what we say into text and failing to keep up with human timing. The AI keeps gaining power and independence, but remains deaf to the emotion in our voices.

This, the team believes, poses a safety problem. Dessa is a step towards addressing this issue. The Dessa team has also stated that it will provide "a small group of AI safety researchers" with early access to the model.