In the heart of Shanghai, an AI start-up is making waves and betting on the transformative power of "scaling law" to boost its AI capabilities. Founded by Jiang Daxin, a former vice-president at Microsoft and a prominent figure in AI research, StepFun, or Jieyue Xingchen in Chinese, is forging a path towards AI supremacy despite facing significant challenges.
A Bold Vision Amid Challenges
StepFun's ambitious goal is to harness the scaling law, which posits that as the size of an AI model and its training data increase, the model's performance enhances exponentially. Zhu Yibo, head of systems at StepFun, underscored this vision at a recent media briefing in Shanghai, emphasizing the relentless demand for higher computing power in China amid the generative AI boom. "Computing power, systems, data, and algorithms are the cores in the pursuit of the scaling law," Zhu declared, highlighting the foundational elements driving their innovation.
Pushing Boundaries with Step-1V and Step-2V
Since its inception in 2023, StepFun has been on a relentless pursuit of AI excellence. The start-up has already launched the Step-1V multimodal large language model (LLM), boasting over 100 billion parameters. But they aren't stopping there. Currently, StepFun is testing the Step-2V model, which features a staggering one trillion parameters, promising unparalleled performance and capabilities.
Navigating Chip Restrictions
One of the significant hurdles for Chinese AI start-ups like StepFun is the restricted access to advanced AI chips from U.S. suppliers like Nvidia. These chips are crucial for achieving the high computing power required for large-scale AI models. However, Zhu remains optimistic, describing these challenges as "manageable," though he did not delve into specifics.
A Powerhouse Team
StepFun's meteoric rise in China's AI scene is partly attributed to the stellar backgrounds of its founders. Jiang Daxin's illustrious 16-year career at Microsoft saw him spearhead groundbreaking projects such as the Bing search engine, the intelligent voice assistant Cortana, Azure cognitive services, and natural language systems for Microsoft 365. His co-founders, Zhu Yibo and Jiao Binxing, also bring valuable experience from their tenures at Microsoft, further strengthening StepFun's leadership.
Building a Future-Ready Computing Center
StepFun is not just about big ideas; they are making substantial investments in infrastructure. The company's computing center, currently under development in Shanghai, is poised to become one of China's premier AI facilities. This move is a testament to StepFun's commitment to providing the necessary computing power to support their ambitious AI models.
China's AI Ecosystem: A Competitive Arena
China's domestic AI market is teeming with innovation, with over 200 large language models (LLMs) developed by various companies. Major players like Zhipu AI, Baichuan, Moonshot AI, and Minimax claim to have some of the top-performing models globally. Yet, the commercial viability of these advanced AI products remains an open question, highlighting the competitive and uncertain nature of the market.
Conclusion
StepFun's audacious bet on scaling law underscores its commitment to pushing the boundaries of AI. Despite the challenges posed by U.S. export controls and fierce competition within China, StepFun's cutting-edge models and the expertise of its founders position it as a formidable player in the global AI landscape. As the company continues to innovate and expand its capabilities, it is poised to play a pivotal role in shaping the future of AI.