Tian Yuandong, a distinguished researcher specializing in deep learning, shared insights into his academic trajectory, his experience in conducting research on Go AI and reinforcement learning at Meta FAIR, and his current entrepreneurial endeavors during a podcast interview. He revealed that after collaborating with GPT-5 to co-author a paper, his work efficiency skyrocketed, multiplying several times over. Tian expressed the belief that his research endeavors could be supplanted by AI within the next five years, a realization that has spurred him to embark on an entrepreneurial venture focused on recursive superintelligence. He emphasized that while AI can significantly boost research efficiency, it still falls short of independently generating high-quality, original ideas. Human oversight remains essential for problem selection and result validation. Furthermore, Tian discussed the challenges hindering progress in technical areas such as reinforcement learning, highlighting that new architectures must be grounded in clear, scalable core insights to overcome scalability bottlenecks. He also voiced opposition to stringent restrictions on recursive self-improvement, advocating for open-source research to prevent the monopolization of top models by a select few. Additionally, he predicted that open-source models will soon rival the usability of their closed-source counterparts.