Say goodbye to robotic, walkie-talkie-style AI voices. On today’s episode, we dive into the open release of NVIDIA PersonaPlex-7B-v1, a speech-to-speech model that can actually hold a real back-and-forth conversation. Built for full-duplex communication, PersonaPlex listens and talks at the same time, handles interruptions, and brings natural rhythm to AI-driven dialogue. We break down why this model is a game-changer for creators, live hosts, and anyone building audio products. Discover how PersonaPlex moves beyond clunky handoffs between old-school speech-to-text and large language models, and nearly erases awkward pauses. Hear tips on setting up creative “producer goblin” AI co-hosts, why backchanneling is both a superpower and a potential dark pattern, and how brands can avoid the “smug millennial barista” voice trap. We also spotlight creator-friendly use cases, from improv livestreams to real-time chat recaps, and discuss what open weights mean for teams of all sizes. Plus, stay tuned for a rapid-fire roundup on Runway Gen-4.5, Adobe Firefly’s latest integrations, and OpenAI ChatGPT Atlas’s new tab-wrangling tricks. If you want to know where conversational AI and creator tools are headed, this episode puts you right at the center of the action.