Block Unveils AI Model That Chats Like an Awkward New Friend

Jack Dorsey’s Block has unveiled an AI model named Dessa, which has been developed entirely from human conversations. To demonstrate its capabilities, Block shared some of the interactions Dessa has had with users, showcasing moments that can be both awkward and intriguing.

For instance, in a conversation highlighted by Dessa’s lead, Ragavan Thurairatnam, the model engages in a mostly fluid exchange, although it amusingly confuses the speaker with someone named “Ronaldo” at the start.

Dessa also participated in a voice-based guessing game, which you can check out in a video shared by the team.

The Dessa website showcases various examples, including a discussion about Christopher Nolan’s films, where Dessa unexpectedly veers into a tangent about Bryan Singer’s Superman movies. Another interaction features a user sharing their surprise trip to Paris, prompting Dessa to respond with what seems like heartfelt gratitude. Additionally, there’s a conversation between Dessa and OpenAI’s GPT-Live-1 voice model, where Dessa exhibits a more natural conversational style. Clips from these exchanges can be found in another video shared online.

According to the Dessa team, this model is a “true speech-to-speech model,” with all its knowledge derived from human speech. It can operate locally on a MacBook Pro M4 Max or better. The team outlines their motivations for creating Dessa as “curiosity (aka fun), better understanding between humans and AI, and safety,” providing further insights:

> Babies learn to talk without ever seeing a word written down. We wanted to see if a model could too. It can. It’s just very hard.
> Speech is the most natural interface to any intelligence, super or otherwise, and today’s voice AI wastes it. It flattens what we say into text and can’t keep up with human timing.
> AI keeps gaining power and independence, and it’s still nearly deaf to the emotion in our voices. Imagine being in serious pain while a voice AI has no idea and just keeps going. That’s the safety problem. Dessa is a step toward fixing it.

The Dessa team plans to provide early access to the model for a select group of AI safety researchers.