Block, the company founded by Jack Dorsey, has introduced Dessa, an AI model designed to engage in conversations that feel remarkably human. The model is trained exclusively on human speech, aiming to replicate the natural flow of dialogue. Dessa's developers showcased several examples of its interactions, including a conversation where it initially misidentifies a user as 'Ronaldo'.
The Dessa website features a range of examples, such as discussions about Christopher Nolan's films and a touching exchange about a surprise trip to Paris. In one instance, Dessa responds with what sounds like emotional gratitude, highlighting its ability to mimic human sentiment. The model also interacts with OpenAI's GPT-Live-1 voice model, where it appears more natural in conversation.
According to the Dessa team, the model is a 'true speech-to-speech' system, capable of running on a MacBook Pro M4 Max or better. The development of Dessa was driven by three main goals: curiosity, improving human-AI understanding, and enhancing safety. The team emphasized that speech is the most natural interface for communication, and current voice AI systems often fail to capture the nuances of human interaction.
The project also addresses the safety aspect of AI, as current systems struggle to interpret emotional cues in speech. Dessa aims to bridge this gap by creating a more responsive and emotionally aware AI. The team plans to provide early access to a select group of AI safety researchers to further refine the model and explore its potential applications.
Overall, Dessa represents a significant step forward in AI conversation technology, offering a more natural and emotionally attuned experience. Its development underscores the growing focus on creating AI that can better understand and respond to human emotions and nuances in speech.