OpenAI has rolled out new voice models for ChatGPT that finally let the AI listen and talk at the same time - a skill many humans still struggle with. Dubbed GPT-Live, the upgrade uses a "full-duplex architecture" to interject with thoughtful "Yeah" and "Mhmm" noises while you ramble, making it feel almost like talking to a real person who's just pretending to pay attention.

Available from Wednesday on the ChatGPT website, Windows app, and mobile apps, GPT-Live comes in two flavors: GPT-Live-1 for paying customers (Go, Plus, Pro) and GPT-Live-1 mini for the freeloaders. The mini model uses fewer resources, which is a polite way of saying it's the budget version of pretending to care. You can switch between Live, Advanced, or Standard Voice in Settings, but why would you go back now that your AI can multitask?

In testing, the model handled interruptions like a champ - telling a story about a cat named Mr. Giggles traveling to the moon, then pivoting to Mars mid-sentence without missing a beat. It also managed live translation between English and French, searched the web during a conversation about iPad camera aspect ratios, and discussed classic Hollywood films with the enthusiasm of a film student who's had too much caffeine. One ZDNET editor noted that background noise occasionally derailed the chat, but your mileage may vary - just turn down the TV or shush your family.

With this update, ChatGPT feels less like a stilted robot and more like that friend who lets you interrupt them constantly. Whether that's a good thing depends on how much you like hearing yourself talk.