Anthropic Co-Founder Predicts More Significant AI Advances in 2025

Recently, the remarkable achievements of OpenAI's o3 model have once again demonstrated that the development of artificial intelligence (AI) is not slowing down but may be accelerating. This perspective is strongly supported by Jack Clark, co-founder of Anthropic, who elaborated on it in his widely-read newsletter, "Import AI."

Clark firmly refuted the notion that AI development is nearing its limits. He stated, "The claims that AI progress is slowing or that expansion is hitting a bottleneck do not align with the current reality." He emphasized that OpenAI's new o3 model is a powerful testament to the vast potential for growth in AI, which is not solely dependent on increasing model size but also on the integration of reinforcement learning and more robust computational power.

Clark highlighted that the o3 model's capability to "think aloud" during operation introduces new possibilities for AI expansion. He predicted that by 2025, this trend will accelerate, with companies combining traditional methods (such as scaling base models) with new approaches in training and inference, driving further advancements in AI technology.

However, Clark noted that despite the promising future of AI, most people are not adequately prepared for the rapid changes to come. He warned, "I believe that almost no one can fully predict the magnitude of AI progress from now on." This viewpoint has sparked extensive discussions and attention regarding the future of AI.

In terms of computational costs, Clark identified a significant challenge. He explained that the most advanced version of o3 requires 170 times the computational power of its basic version, far exceeding the previous o1 model, which itself required more computational resources than GPT-4. These new systems make cost predictions more complex and difficult, as resource demands vary significantly depending on the specific tasks.

Despite these challenges, Clark remains optimistic about the future of AI. He predicts that combining traditional scaling methods with new approaches will lead to even more "remarkable" AI advancements in the coming years. This view is widely shared and anticipated in the industry.

Additionally, Clark's predictions have sparked speculation about Anthropic's future directions. While the company has not yet released a "reasoning" or "test-time" model to compete with OpenAI's o series or Google's Gemini Flash Thinking, it is generally believed that Anthropic will continue to play a crucial role in advancing AI technology.

As AI technology continues to develop and expand into new application areas, expectations for the future of AI are growing. Clark's insights and forecasts inject new energy and confidence into the field, heightening anticipation for future innovations and breakthroughs in AI technology.