The latest upgrade to ChatGPT, a conversational AI chatbot, is a significant step forward in voice interaction capabilities. The new voice mode, dubbed GPT-Live, is being rolled out to all users and promises to make conversations feel more natural and fluid.
GPT-Live is a "new generation" of voice models that combines the ability to think in the background while continuing the conversation. This allows users to engage with the AI in a more natural way, with extra space to pause when needed and indicators that the AI is still listening. The upgrade is available on mobile and web platforms now, or very soon.
The previous voice mode in ChatGPT was a relatively basic wrapper on top of the standard text input and output. GPT-Live, on the other hand, uses a more advanced approach that allows the AI to be talking and listening at the same time. This enables faster and more nuanced responses, as well as improved recognition of when a user is pausing mid-thought or finishing a query.
One of the key benefits of GPT-Live is its ability to handle complex tasks in the background while the conversation continues. This means that users can ask the AI to take a beat or slow down, and even perform tasks like translating text into a foreign language in real-time. Visual responses have also been improved, with pop-up cards for locations, weather forecasts, and sports scores.
To access the new voice mode, users can tap the soundwave-style icon on the mobile app. The upgraded voice mode feels more realistic and natural, with the AI talking in a varied and expressive way. ChatGPT will also hesitate and draw words out at times, making the conversation feel more like a real conversation.
There are three "intelligence" levels to choose from, with varying levels of trade-off between the speed of the response and how detailed and accurate it is. Users can tweak the level of thinking ChatGPT puts into the new voice mode by tapping the sliders icon and selecting the desired level.
While there were a few moments where the speech glitched and the conversation hung, these bugs are likely to get ironed out over time. Interruptions are handled well, with the AI pausing to acknowledge what the user has said and then continuing its train of thought.
