Towards Safe, Calibrated and Accessible Artificial General Intelligence

National AI Research Lab (NAIRL) hosted an invited talk by Pascal Poupart, Professor at the David R. Cheriton School of Computer Science at the University of Waterloo, on October 29, 2024, at KAIST Kim Jaechul Graduate School of AI on KAIST Seoul Campus.

Under the theme “Towards Safe, Calibrated and Accessible Artificial General Intelligence,” Professor Poupart noted that recent advances in AI have led to the emergence of generalist agents with astounding capabilities, spanning language, image and video generation, autonomous driving, complex reasoning, and advanced decision making. He presented recent developments in inverse constraint reinforcement learning for learning safety constraints from human demonstrations, and discussed uncertainty quantification techniques that help agents know what they do not know. He also introduced recent advances in reward guided image and text generation, which reduce the training costs of diffusion models for subject driven image generation and modularize LLM alignment without prohibitively expensive fine tuning, improving the accessibility of large language models.

Professor Poupart is a Canada CIFAR AI Chair at the Vector Institute and a member of the Waterloo AI Institute. He serves on the advisory board of the NSF AI Institute for Advances in Optimization, and previously served as Research Director and Principal Research Scientist at the Waterloo Borealis AI Research Lab at the Royal Bank of Canada. His research focuses on machine learning algorithms with applications to natural language processing and material discovery, and he is best known for his contributions to reinforcement learning. His team’s ongoing projects include democratizing large language models, inverse constraint learning, mean field RL, Bayesian federated learning, uncertainty quantification, and material discovery for CO2 recycling. NAIRL will continue to connect Korea’s research community with leading AI experts from around the world through its invited talk series.