Recently, another significant piece of news in the field of artificial intelligence has drawn attention, coming from Jack Clark, co-founder of Anthropic. In his newsletter "Import AI," he refuted earlier claims about a slowdown in AI progress and pointed out that AI development is actually accelerating. Clark mentioned that the recently launched o3 model by OpenAI indicates that there is still substantial room for growth in AI, although the approach to achieving this is changing.
Unlike traditional methods, the o3 model does not merely push progress by increasing the model's size; instead, it employs reinforcement learning and more powerful computing capabilities. During the model's operation, o3 possesses the ability to "think while running," which opens up entirely new possibilities for further expansion. Clark predicts that this trend of combining traditional methods with emerging technologies will accelerate by 2025, when more companies will integrate large foundational models with new computing approaches to drive AI advancements.
However, Clark also noted a significant challenge behind this rapid development—computing costs. He stated that the computational power required for the advanced version of o3 is 170 times that of the basic version, while the resource consumption of the basic version has already surpassed that of o1, and o1 itself has a higher energy consumption than GPT-4o. The resource demands of these new systems vary depending on the tasks, making cost predictions increasingly complex. In the past, the expenses associated with models were primarily related to their size and output length, but the flexibility of o3 complicates this predictability.
Nevertheless, Clark firmly believes that by combining traditional scaling methods with new approaches, AI development in 2025 will be more significant than ever. His predictions regarding Anthropic's future plans have garnered widespread attention. Currently, Anthropic has not released any "reasoning" or "test-time" models to compete with OpenAI's o series or Google's Gemini Flash Thinking. The company's flagship model Opus3.5 has been temporarily shelved due to insufficient performance improvements; however, the development of this model has not been a complete failure and has played a crucial role in training the new Sonnet3.5 model, which has become the most popular language model on the market.
Key Points:
🌟 AI development is not slowing down; instead, it is accelerating, especially with models that combine traditional and emerging computing methods.
💡 The o3 model's "think while running" capability opens up new expansion possibilities, injecting momentum into the future of AI.
💰 Despite rapid progress, the uncertainty of computing costs remains a significant challenge for future development.
