One of the most stubborn problems in conversational voice AI, that has been for decades… is figuring out when someone is actually finished talking. It sounds trivial, but every voice assistant you've ever used, Alexa, Siri… etc. has had to deal with this problem. The simplest solution has always been to just detect silence. But if the pause you wait for is too short, you cut people off mid-thought; and if you stretch that pause out, the whole thing starts to feel sluggish and dumb, sitting there waiting after you've clearly finished a simple sentence. It's something we do so intuitively as humans, yet AI systems continue to struggle with it, and it goes to show how much latency, and knowing when to jump in, contributes to a conversation actually feeling natural. This update to GPTARS is just one of a handful aimed at making conversations with him feel more natural, and it's something I've been working on for quite a while. I figured now was a good time to start showing some of these updates, since @OpenAI is also rolling out their new live voice update to @ChatGPT with the same goal of enabling more natural conversations. Don't get me wrong, what they're doing is leaps and bounds more technically complex than what's going on here. I just found it interesting how it highlights some of the obvious shared problems in this space, and how we're all chasing the same result, albeit in very different ways of getting there. If you’re interested in the other updates I have been making in pursuit of this, make sure you follow. #ai #robot #chatgpt #technology #openai - @gptars