Jack Dorsey's New AI Talks Like a Friend, Not a Robot

Jack Dorsey's company Block has built a new AI model called Dessa that learns exclusively from human speech, not written text. This is unusual because most AI models, like ChatGPT, are trained on massive amounts of text from the internet. Dessa, by contrast, was trained 'from scratch on nothing but human conversation.' The goal is to create a voice AI that sounds more natural and can pick up on emotions in our speech.

In demonstrations, Dessa has conversations that are mostly natural but occasionally awkward—like a new friend who's still learning social cues. For example, it mistakenly called someone 'Ronaldo' at first. But it also showed emotional responses, like sounding teary-eyed when told about a surprise trip to Paris. The model is a 'true speech-to-speech' system, meaning it processes spoken words directly without converting them to text first. This allows it to keep up with human timing and tone.

Block says it built Dessa for three reasons: curiosity, better understanding between humans and AI, and safety. The safety concern is that today's voice AI often ignores emotional cues—imagine being in pain while a voice assistant keeps chatting cheerfully. Dessa is a step toward AI that can detect and respond appropriately to human emotions. The model can run locally on a MacBook Pro M4 Max or better, meaning it doesn't need cloud servers.

For now, Dessa is being given to a small group of AI safety researchers for early access. While it's not yet available to the public, it represents a shift toward more emotionally intelligent voice assistants that could one day help with tasks like customer service, healthcare, or even companionship. The catch? It's still experimental and may have awkward moments, but it's a promising move toward AI that truly understands us.