O1 by OpenAI: A New Paradigm in AI?

In the early hours of Friday, September 13, Korea time, OpenAI quietly introduced a new AI model—O1. Formerly known by the codename "Strawberry" and once rumored as "Q*" (sparking AGI-related conspiracy theories),

In the early hours of Friday, September 13, Korea time, OpenAI quietly introduced a new AI model—O1. Formerly known by the codename "Strawberry" and once rumored as "Q*" (sparking AGI-related conspiracy theories), O1 has been surrounded by intense anticipation. This model showcases a significant leap in AI reasoning, notably in its use of the "Chain of Thought" process. But beyond just another powerful AI, OpenAI has called O1 a "new paradigm" in AI technology. What does that mean?

Traditional AI models, such as GPT, operate under the "Law of Scale," which suggests that increasing the parameters and data available to an AI model improves its performance. This has led to the scaling up of models and large computing requirements, driving up demand for hardware like NVIDIA GPUs.

However, O1 breaks from this tradition. Instead of just expanding the number of parameters, O1 enhances performance by giving the AI model time to "think" and learn. Unlike previous models that provide an answer after processing a vast amount of data in one go, O1 engages in an iterative thought process. It reflects and revisits its reasoning, much like a human might when solving a complex problem.

According to Mira Murati, CTO of OpenAI, and Jakub Pachocki, the chief scientist, this shift doesn’t completely replace the Law of Scale but complements it. Users of GPT's premium services might have noticed that O1 and GPT-4o are now available side-by-side, allowing users to choose which model suits their needs best.

The release of O1 signifies a broader shift in AI development. While scale has driven past improvements, new ways of enhancing AI, such as giving models time to learn and think, are emerging as alternatives. This dual approach could change how AI models are developed and applied in real-world scenarios.

OpenAI's O1 is more than just another step in AI technology—it represents a shift in how AI is conceptualized and developed. By moving away from purely scaling data and parameters to encouraging models to "think," we might be witnessing the birth of a new era in AI performance optimization. As this paradigm continues to evolve, it will be fascinating to see how it reshapes both AI research and real-world applications.