How ChatGPT becomes an interpreter

The day chosen was everything, but certainly not a coincidence. Just 24 hours before Google's long-awaited I/O developer conference, OpenAI, creator of the artificial intelligence (AI)-based chatbot ChatGPT, invited people to a somewhat spontaneous presentation. He was presented in the introduction by OpenAI CTO Mira Moratti.

A new AI model GPT-4o is introduced, where the “o” stands for “omni”. Which means that the model works multimodally. Both aspects, i.e. both in the input and in the subsequent creation of your content. But this time, the opportunity to talk to ChatGPT over GPT-4o received a particularly positive response. Until now, you could control your chatbot only via voice. If you did this, three AI models worked together: one translated language into text, a second responded to the text with text, and a third converted the response text into language again. A process that led to seconds of waiting after the question and complicated the conversation.

One model is enough

With GPT-4o, where only one neural network is running, the spoken answer must now occur within 300 milliseconds. Meanwhile, the AI's voice has taken on a more human tone. During the demo itself, OpenAI had plenty of application examples for the new AI. So ChatGPT told dramatic bedtime stories, served as an Italian and English interpreter, and even started singing.

German technology expert Philipp Kloeckner talks about a potential “Siri killer” in a simulation of Apple's voice assistant. OpenAI itself is trying hard to draw comparisons to Spike Jonze's 2013 hit film “Her.” Shy Jonathan falls in love with Samantha, the voice of a particularly sympathetic AI.

It is still unclear when the new functions will actually be available. He is currently speaking in voice mode According to OpenAI President Sam Altman Not yet with GPT-4o, while the new model is already being written.