At the KDDI Summit, OpenAI Japan CEO opens up about GPT-Next, an AI model 100 times more powerful than GPT-4. This model, hailed as the successor to GPT-4, promises significant advancements in AI functionality, including smarter training models and enhanced multimodal capabilities.
At the KDDI Summit 2024, OpenAI Japan CEO, Tadao Nagasaki, revealed that GPT-Next will be 100 times more powerful than GPT-4. According to Nagasaki, GPT-Next’s improvements will come not from vast computing resources but from better model architecture. This marks a leap in the AI’s performance without the need for exponential resource scaling.
GPT-Next will incorporate a smaller version of OpenAI’s mysterious Project Strawberry, also known as Project Q. This project is believed to be OpenAI’s effort to create Artificial General Intelligence (AGI). While the full potential of Project Strawberry is still being explored, GPT-Next will leverage some of its capabilities, pushing the boundaries of what AI can achieve.
Rapid Growth of ChatGPT
OpenAI Japan CEO opens up about GPT-Next, an AI model 100 times more powerful than GPT-4, highlighting its efficiency in handling multimodal data like text, images, and audio. Nagasaki also shared impressive growth statistics at the summit. By August 2024, ChatGPT had surpassed 200 million active users, making it the fastest-growing software in history. He emphasized that AI’s accessibility through tools like ChatGPT is making it more integrated into everyday life.
OpenAI’s Chief Technology Officer, Mira Murati, had previously introduced GPT-4o, an upgraded version of GPT-4. This model was already twice as fast, 50% cheaper, and had five times higher rate limits than the GPT-4 Turbo. Murati described GPT-4o as a pivotal moment for human-AI interaction, with enhancements in text, vision, and audio capabilities.
Looking ahead, Murati hinted that GPT-Next might surpass human intelligence in specific tasks. Speaking at Dartmouth Engineering, she likened GPT-3’s intelligence to that of a toddler, while GPT-4 was compared to a high school student. With GPT-Next, she suggested AI could achieve PhD-level intelligence in specialized areas, raising the possibility of AI outperforming humans in certain tasks.
A Glimpse Into the Future
During a recent event, OpenAI Japan CEO opens up about GPT-Next, an AI model 100 times more powerful than GPT-4, emphasizing its improved model architecture for faster processing. OpenAI’s ambitious projects like GPT-Next and Project Strawberry are part of the company’s broader mission to push the limits of AI.
Though GPT-5 is not expected soon, the next phase of AI, driven by GPT-Next, could redefine the way people interact with technology. The journey toward AGI continues to gain momentum, with each model bringing AI closer to human-like intelligence.
System 2 Thinking to Transform AI Problem Solving
OpenAI’s GPT-Next introduces a groundbreaking shift in artificial intelligence with the implementation of “System 2 thinking.” Unlike previous models that rely heavily on predicting token sequences to generate responses, GPT-Next will have the capability to deliberate and reason through problems. This method allows the model to carefully evaluate situations before providing answers, moving away from the traditional prediction-based mechanism. Reuters reports that the introduction of System 2 thinking has already delivered impressive results, with GPT-Next scoring over 90% on the MATH benchmark, a collection of complex mathematical problems.
A significant advancement in GPT-Next is its integration with a scaled-down version of OpenAI’s “Project Strawberry.” This system will not only enhance the model’s computational efficiency but also generate high-quality synthetic training data. One of the primary challenges in AI development is the lack of diverse, high-quality data needed to train advanced models.
By addressing this issue, Project Strawberry is expected to give GPT-Next a major boost in performance, improving its training capabilities while reducing dependence on vast computing resources. This step forward paves the way for even more sophisticated AI models like Orion in the future.
