So I joined a Spaces chat that ended up being a demo of ChatGPT Advanced Voice Mode. The host got access to the alpha and was prompting it live to show functionality. Really he was trying to break it or uncover the inefficiencies. He brought others up to also give voice prompts to see what the output would be. Things like foreign language to foreign language translation (i.e. translation from Spanish to Russian) proved to have challenges but it responded accurately to English translations and prompts. They were telling it do things like make up a story and tell it without narrating, only focus on speaking from the character’s perspective or count from 1-50 in Russian super fast without taking a breath. You could actually hear the model running out of breath trying to deliver that prompt. And they would interrupt the output in real time with a new voice prompt and we got to witness the model change its output in real time to adjust to the new instruction. It was very interesting to hear and watch!