The artificial intelligence industry is at a significant turning point: Leading companies are shifting their development focus from pursuing larger-scale language models to enhancing the thinking capabilities of these models. This transformation will reshape the entire AI industry landscape.

According to Reuters, major AI laboratories are facing difficulties. Developing large-scale language models not only requires tens of millions of dollars in investment but also often encounters technical challenges such as system crashes, with performance evaluations often taking months to complete.

This development bottleneck has affected industry giants. Reports suggest that OpenAI's new Orion model has limited improvements over GPT-4, and Google's Gemini2.0 is facing similar issues. Anthropic's CEO, Dario Amodei, has indicated that the development roadmap for Opus3.5 is being reconsidered.

Artificial Intelligence Robotic Arm AI (2)

Image source: Image generated by AI, authorized by Midjourney

Ilya Sutskever, former co-founder of OpenAI and current head of Safe Superintelligence (SSI), noted, "The 2010s were the era of expansion, and now we are entering a new phase of exploration and discovery." This statement is particularly noteworthy as Sutskever was once an advocate of the "bigger is better" philosophy.

The new industry direction points towards "compute during testing," which involves giving AI models more time to think and solve problems step-by-step. This approach focuses on cultivating the reasoning capabilities of AI systems, enabling them to generate multiple solutions and evaluate them, rather than simply providing quick answers.

This shift could also impact the hardware market landscape. Although Nvidia dominates the traditional AI training hardware sector, the new computing paradigm presents opportunities for other chip manufacturers like Groq. However, the industry anticipates that a combination of traditional and new methods may be adopted in the future to achieve optimal cost efficiency.

Several industry insiders believe that while traditional language model development will continue, the industry's focus has already shifted. This marks the entry of AI development into a new phase that emphasizes quality and thinking capabilities.