Editor's introduction: Yang Zhilin is included in the 'Vision' column not merely because he founded a unicorn company, but because at a critical juncture in the technology cycle, he chose to transform academic frontiers into industrial practice. When the large model wave first emerged, most were still watching; he, however, leveraged his academic work on Transformer-XL and XLNet to quickly enter the niche of long-text processing, using Kimi to demonstrate the intersection of technical depth and product experience. His path represents a typical paradigm among Chinese AI entrepreneurs—the 'academic school'—not chasing trends but positioning ahead based on a deep understanding of technology. This ability to convert academic insight into business judgment is precisely the archival value that the Vision column seeks to preserve.

Yang Zhilin's entrepreneurial journey began in early 2023, when the global AI fever ignited by ChatGPT showed him the window for large language models moving from lab to mainstream. He judged that the Chinese market urgently needed excellent localized solutions, so in April 2023 he founded Moonshot AI. The company name symbolizes a 'moonshot' ambition, representing relentless pursuit of ambitious goals. He assembled a core team of technical talent from Tsinghua, CMU, and other top institutions, combining academic depth with engineering capability. ^[2] In October 2023, Moonshot AI launched Kimi, an intelligent assistant whose core differentiation lies in ultra-long context processing—supporting a lossless context window of 2 million Chinese characters, allowing users to input entire books, complete codebases, or large document sets for understanding, analysis, and summarization. ^[2] Kimi quickly gained popularity, with registered users surpassing 10 million within months and daily active users reaching millions. ^[3] In the capital market, Moonshot AI also performed strongly, completing funding of several hundred million USD in 2024 with a post-money valuation exceeding $3 billion, making it one of the fastest-growing AI startups globally. ^[3]

Yang Zhilin's technology philosophy is rooted in his academic research. During his undergraduate studies at Tsinghua University's computer science department, he demonstrated extraordinary talent in AI, publishing multiple high-level papers in natural language processing. ^[1] During his PhD at Carnegie Mellon University under Professor Tom Mitchell, he deeply researched language model core architectures and training methods. ^[1] His representative research contributions include Transformer-XL and XLNet. Transformer-XL solved standard Transformer's context fragmentation in long-sequence modeling by introducing recurrence mechanisms, laying technical foundations for subsequent long-text processing; XLNet overcame BERT-type models' pre-training and fine-tuning inconsistency through permutation language modeling, achieving leading results on multiple NLP benchmarks. ^[1] These research results directly influenced Kimi's ultra-long context processing capability. In engineering implementation, the Moonshot AI team innovatively adopted techniques such as grouped query attention (GQA) to greatly extend context length while maintaining inference speed. ^[2] Yang Zhilin adheres to the philosophy of 'technology-driven, product-first', believing that the core competitiveness of large model companies lies in delivering genuinely valuable product experiences to users, not just benchmark rankings. ^[3]

Yang Zhilin and his Moonshot AI have had a profound impact on China's AI field. The launch of Kimi allowed ordinary users to experience for the first time the powerful capabilities of large models in long-text understanding, and it has been widely used in scenarios such as paper reading assistance, legal document analysis, long-document summarization, and code debugging. ^[2] Moonshot AI's technical strength has been widely recognized in the industry; its self-developed MoE (Mixture of Experts) architecture model has performed excellently in multiple evaluations, especially in long-text understanding and reasoning capabilities, leading the industry. ^[3] In the capital market, Moonshot AI has received investment from well-known institutions such as Sequoia Capital China and ZhenFund, with investors generally optimistic about Yang Zhilin's team's technical strength and commercialization ability. ^[3] Yang Zhilin's ultimate goal is to achieve Artificial General Intelligence (AGI), which he believes is not a distant sci-fi concept but a technical goal that can be gradually realized in the foreseeable future. ^[3] His story is a microcosm of China's new generation of AI entrepreneurs—possessing top academic backgrounds, global vision, and unwavering belief that technology can change the world. ^[4]

The business proposition Yang Zhilin leaves behind is: in a technology-driven industry, how does academic depth translate into product competitiveness? He has demonstrated that when a technology inflection point arrives, teams with underlying innovation capabilities can seize opportunities faster. However, the exploration of AGI is far from over, and whether Moonshot AI can maintain its lead in fierce market competition remains an open question. His archival value lies in documenting how a young entrepreneur, starting from academia, found his coordinate in the business world and provided a possible path for the development of AI in China.