The field of artificial intelligence is undergoing a profound change, and the scale competition of large-scale language models has gradually given way to improvements in model thinking ability. Major AI laboratories have adjusted their development strategies, shifting from pursuing model scale to improving their ability to reason and problem-solving, which indicates that the AI industry will enter a new stage of development. This article will explore in-depth the reasons behind this transformation, specific manifestations and its impact on the future development of the industry.
The artificial intelligence industry has ushered in a major turning point: leading companies have changed their development direction, from pursuing larger-scale language models to focusing on improving their thinking ability. This transformation will reshape the development pattern of the entire AI industry.
According to Reuters, major AI labs are facing difficulties. Developing large language models not only requires tens of millions of dollars, but also often encounter technical difficulties such as system crashes. Evaluating the performance of a model often takes months.
This development bottleneck has affected industry giants. There are reports that OpenAI's new Orion model has limited improvements compared to GPT-4, and Google's Gemini2.0 has also encountered similar difficulties. On the one-year-old Anthropic, its CEO Dario Amodei said it is re-planning the development route of Opus 3.5.
"The 2010s are an era of expansion, and now we are entering a new stage of exploration and discovery," said Ilya Sutskever, former co-founder of OpenAI and now head of Safe Superintelligence (SSI). Advocate of the concept of "bigger, better".
The new industry direction points to "test-time computing", that is, giving AI models more time to gradually think about and solve problems. This approach focuses on developing the reasoning capabilities of AI systems, enabling them to generate multiple solutions and evaluate them instead of simply answering quickly.
This change may also affect the hardware market structure. While Nvidia dominates the traditional AI training hardware space, the new computing paradigm presents opportunities for other chip manufacturers such as Groq. However, the industry expects that both traditional and new approaches may be adopted in the future to achieve optimal cost-effectiveness.
Many industry insiders believe that although the development of traditional language model will continue, the focus of the industry has begun to shift. This marks the entry of a new stage of AI development that focuses more on quality and thinking ability.
The AI industry's shift to model-oriented thinking ability marks a new stage in its development. This transformation will not only affect the development direction of the model, but will also have a profound impact on the hardware market and the entire AI industry structure. Future development is worthy of continuous attention.