Summary
Andrew Bai is an AI scientist and fifth-year PhD candidate at UCLA with nine years of industry and research experience focused on memorization and forgetting in machine learning, particularly in post-training for LLMs. He has helped demystify why RLHF reduces forgetting compared to supervised fine-tuning and developed practical methods—like computationally-free rehearsal and early-stopping metrics—that improve alignment efficiency and downstream DPO performance. His internships at DeepMind, Google, and Amazon produced high-impact prototypes and benchmarks (e.g., a tool-calling prototype with sub-1s latency and a pedagogically inspired response-format benchmark), showing he bridges rigorous research with production-minded systems. Comfortable across reward modeling, diffusion memorization, long video generation, and LLM agents, he thrives on collaborative problem solving and cross-disciplinary projects. Based in Los Angeles and now at Mistral AI, he pairs deep theoretical insight with hands-on engineering to make alignment research more scalable and practical.
9 years of coding experience
3 years of employment as a software developer
Bachelor's degree, Computer Science, Bachelor's degree, Computer Science at National Taiwan University
University of California, Los Angeles
English, Chinese