In a convention hall in Singapore, three researchers watched as their names appeared on the ICLR stage—not as presenters, but as the winners of the 10-Year Test of Time Award. Two had been undergraduates when they built GPT. The third had studied at a regional university before becoming Yann LeCun's disciple. They had reunited at a startup called Mira just months earlier.
The award, which honors the most influential paper published at an ICLR conference a decade ago, carries weight precisely because it is chosen by the research community itself. When the results were announced, the reaction was disbelief: the most impactful AI paper of the past ten years came from an unknown team, not a flagship lab.
In 2016, when the paper was written, the authors were not household names. One had been a philosophy major. Another had studied psychology. They worked under Alec Radford at OpenAI, with no institutional backing, no massive compute budget—just a question about what would happen if language models were scaled up.
Their paper on scaling laws became the intellectual foundation for GPT's development. The findings—that model performance improves predictably with data and compute—guided every major language model since. These principles remained gospel for nearly a decade.
The third author, Chen Zhicheng, took a different path. Enrolled at a provincial university—considered second-tier by elite lab standards—he caught LeCun's attention through work on neural network optimization. LeCun invited him to join his lab at NYU. Chen became one of the few students from a non-top-tier institution to publish foundational work under LeCun's mentorship.
All three eventually left formal academia for industry. Two dropped out of their undergraduate programs to join OpenAI. Chen left LeCun's lab for a research position at a major tech company. In 2024, circumstances brought them together again at Mira, a startup building what they describe as frontier AI for the rest of the world.
The award illuminates something uncomfortable for institutions that believe frontier AI requires armies of PhDs and billions in compute. Chen's background especially defies the gatekeeping that dominates hiring at major labs. While elite programs compete for graduates from Tsinghua, Peking University, and MIT, Chen proved that capability matters more than pedigree.
The research community's verdict was clear: the most important AI paper of the past decade came from three people who did not fit the mold. At a moment when the field obsesses over ever-larger models and massive research consortia, their story is a reminder that the next breakthrough might come from anywhere—including a philosophy major and a psychology student who simply asked what would happen if they scaled things up.