At the AI Investment Summit held in Beijing on September 16, themed "Deterministic Opportunities in the AI Infrastructure Era," Yu Gang, Vice President of StepFun and Head of On-Device Models, participated in a panel discussion, offering insights into the evolving landscape of artificial intelligence.

Yu highlighted that the shift to AI 2.0 represents a significant transformation in the field. Since ChatGPT was launched in 2022, the fundamental change in how people perceive AI can be summarized in one word: generalization. He contrasted this with the AI 1.0 era, where tasks were difficult to transfer between domains. For instance, technology developed for facial recognition required entirely new pipelines to be adapted for license plate recognition—two completely separate approaches. In the era of large models, however, generalization capabilities have improved exceptionally well, and at a rapid pace.

Initially, when ChatGPT was unveiled in 2022, its primary use was for chatting. Over time, users discovered it could write articles, build websites, and complete coding tasks. Most recently, the release of GPT-6 Astra introduced demos that even showcase operations in embodied scenarios, underscoring how large models are becoming increasingly versatile and capable of handling an ever-widening range of tasks.

Beyond broader task generalization, Yu also emphasized a continuous rise in output quality. Take the generation of front-end web pages as an example: early models could only achieve basic functionality, but now there is a growing demand for visual appeal and aesthetics. The latest models are even competing on 3D rendering, with noticeable improvements in texture and overall visual effects. These two major advancements on the model side—wider task generalization and higher output quality—are trends that all model companies must prioritize going forward.

To stay competitive, Yu stressed that enterprises need to accelerate their investments in computational power, talent acquisition, and organizational synergy. Only by moving decisively in these areas can companies ensure their teams operate with greater efficiency and cohesion, enabling them to fully capitalize on the opportunities presented by AI 2.0.