Damn it, Jack's voice AI has exploded! After watching the demonstration, it was a bit shocking. It can understand your emotions, tone, sarcasm, and pauses, and even burst out laughing or naturally interject halfway through your words. If you raise the volume, it will also speak loudly. If you are feeling down, its voice will also change. After looking at it, Jack was trained from scratch using only human dialogue recordings, without any text or pre trained models. In the past, many voice AI systems had to first convert what you said into text and then hand it over to large models for processing, making it easy to lose tone and emotion. Although OpenAI's Realtime can now directly process speech, Dessa is more radical, completely bypassing text in the process of learning language and learning to communicate like a baby by listening to people speak. Block claims that the entire model can be trained using only one set of NVIDIA GB200 NVL72, and can run directly on the MacBook Pro of M4 Max, supporting real-time two-way dialogue, listening and speaking at the same time, without waiting for you to finish speaking.