Seeing, Hearing, and Responding Like Humans
'Direct Confrontation' a Day Before Google's Event
OpenAI, the developer of ChatGPT, unveiled a new AI model on the 13th (local time) that can engage in natural conversations like a real person. This move directly challenges Google just one day before its annual developer conference (I/O).
On this day, Mira Murati, OpenAI's Chief Technology Officer (CTO), revealed 'GPT-4o' (GPT-4o) during a live event.
Unlike previous models that primarily communicated through text, 'GPT-4o' is an AI model that allows users to ask questions and request answers through real-time voice conversations. It can reason and respond not only through text but also through auditory and visual inputs. The 'o' in the new model stands for 'omni,' meaning 'all.'
The response time has also been significantly reduced. GPT-4o's response time is a minimum of 232 milliseconds and an average of 320 milliseconds. OpenAI explained that this is comparable to human response times. The previous model, GPT-3.5, took an average of 2.8 seconds, and GPT-4 took 5.4 seconds to respond. This overcomes reaction delays, enabling real-time conversations that feel like talking to an actual person.
During the demonstration, GPT-4o showcased the ability to speak in various tones, voices, emotions, and styles. It also solved a simple math problem (3x+1=4) through vision recognition. When asked to show the solution method, it explained step-by-step carefully. It sees, hears, and provides answers like a human.
The scene where the protagonist talks with the AI Samantha in the movie 'Her' comes to mind. Sam Altman, OpenAI's Chief Executive Officer (CEO), posted the word 'her' on X (formerly Twitter) after the event ended.
It also offers real-time translation features. OpenAI stated that the 'GPT-4o' model is twice as fast as the existing GPT-4 Turbo and costs half as much. GPT-4 Turbo is the latest version introduced last November. Additionally, the quality and speed for 50 languages, including Korean, have been improved.
It is provided free of charge to all global users, but existing paid users can ask five times more questions than free users. GPT-4o is available starting today, and the AI voice mode will be released within a few weeks.
CTO Murati said, "I think this is the first time we have made really significant progress in ease of use."
OpenAI's announcement today is expected to intensify AI competition among big tech companies. Google’s major annual developer conference is just one day away. Apple is also expected to announce its AI strategy at the annual Worldwide Developers Conference (WWDC) next month.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
