R&D on LLM post-training
About Me
My name is Yang Boming. I am currently a second-year Ph.D. student (D2) at the University of Tokyo. My research primarily focuses on post-training and Reinforcement Learning (RL) to continuously enhance the capabilities of LLMs.
What I Worked On During the Internship
My internship theme was "R&D on LLM post-training." While the broader scope included data quality evaluation and synthetic data generation, my specific focus and responsibility were centered on Reward Modeling. I focused on the development and optimization of Reward Models for LLM post-training. This involved experimenting with different architectures and loss functions to better align model outputs with human preferences, ultimately contributing to the improvement of the overall model performance.
Why I Applied for the Internship
My main motivation was to bridge the gap between academia and industry. I find it deeply rewarding to apply theoretical methods from research papers to real-world industrial models. PFN is undoubtedly one of the top AI companies in Japan, and I felt honored to learn from such a talented group of experts.
What My Days Looked Like During the Internship
The working hours at PFN are very flexible, which allowed me to commute comfortably. A typical day involved deep-focus development in the morning, followed by a pleasant lunch with my mentors or fellow interns—the atmosphere was always very welcoming. I also greatly enjoyed the "club" culture at PFN, participating in the Reading Club and Programming Club, which added a sense of warmth and fun to the professional environment.
What I Gained from the Internship
First, I gained a profound understanding of the difference between lab research and industrial application. While academic research often operates in idealized conditions to prove a theory, industrial systems must handle real-world noise and complexity, which was a highly educational experience. Second, I learned the importance of "asking the right questions." In a fast-paced R&D environment, proactive communication with mentors and peers helped me unblock technical hurdles much faster than working in isolation.
Who Should Apply?
I highly recommend the PFN Internship for those who value a flat organizational structure and a free, open atmosphere for technical discussion. If you are looking for a workplace that combines cutting-edge research with warm, engaging community activities, PFN is the perfect place for you. Please don't hesitate to apply!
