(Title)
Reasoning, Alignment, and Creativity: The Triptych of Superintelligence
(Bio)
Moontae serves as Head of Superintelligence Lab at LG AI Research. He is concurrently a faculty member of Information and Decision Sciences at the University of Illinois Chicago. His journey into Large Language Models (LLMs) began in 2019 as an invited scholar at Microsoft Research Redmond, where he initiated the ambitious Universal Language Modeling project. His current research spans text, code, and time-series foundation modeling, with an industry service on synthesizing high-quality domain-specific reasoning datasets for agent building and verifiable thinking verification. Moontae has served as Area Chair and Senior Committee member for NeurIPS, ICML, ICLR, ACL, NAACL, EMNLP, AAAI, AISTATS, and CVPR. Beyond the machine learning community, his work have also been recognized in Operations Research and Management Information Systems, where he received the Best Paper Award at INFORMS 2017. His research in Computational Social Science won the Amazon Research Award. More recently, he received the Social Impact Award at NAACL 2024 and the Best Paper Award at NAACL 2025.
(Abstract)
The pursuit of superintelligence requires advancing three interdependent pillars: reasoning, alignment, and creativity. Reasoning calls for foundation models that can perform multi-step thinking and workflow modeling supported by scalable methods for structured planning and evaluation. This requires synthesizing domain-specific reasoning datasets, designing protocols for verifiable thinking, and establishing principled evaluation frameworks to ensure reliability and trustworthiness in Agentic AI. Alignment bridges these capabilities with human preferences, requiring rigorous and safe guardrails for societal values and collective goods. Creativity, in turn, expands the frontier—enabling AI not only to interpolate within learned patterns but also to extrapolate toward novel insights. By treating reasoning, alignment, and creativity as a triptych rather than isolated challenges, this talk aims to envision AI systems that are not only more capable and reliable, but also generative in expanding scientific discovery and human knowledge.
(Title)
Reasoning, Alignment, and Creativity: The Triptych of Superintelligence
(Bio)
Moontae serves as Head of Superintelligence Lab at LG AI Research. He is concurrently a faculty member of Information and Decision Sciences at the University of Illinois Chicago. His journey into Large Language Models (LLMs) began in 2019 as an invited scholar at Microsoft Research Redmond, where he initiated the ambitious Universal Language Modeling project. His current research spans text, code, and time-series foundation modeling, with an industry service on synthesizing high-quality domain-specific reasoning datasets for agent building and verifiable thinking verification. Moontae has served as Area Chair and Senior Committee member for NeurIPS, ICML, ICLR, ACL, NAACL, EMNLP, AAAI, AISTATS, and CVPR. Beyond the machine learning community, his work have also been recognized in Operations Research and Management Information Systems, where he received the Best Paper Award at INFORMS 2017. His research in Computational Social Science won the Amazon Research Award. More recently, he received the Social Impact Award at NAACL 2024 and the Best Paper Award at NAACL 2025.
(Abstract)
The pursuit of superintelligence requires advancing three interdependent pillars: reasoning, alignment, and creativity. Reasoning calls for foundation models that can perform multi-step thinking and workflow modeling supported by scalable methods for structured planning and evaluation. This requires synthesizing domain-specific reasoning datasets, designing protocols for verifiable thinking, and establishing principled evaluation frameworks to ensure reliability and trustworthiness in Agentic AI. Alignment bridges these capabilities with human preferences, requiring rigorous and safe guardrails for societal values and collective goods. Creativity, in turn, expands the frontier—enabling AI not only to interpolate within learned patterns but also to extrapolate toward novel insights. By treating reasoning, alignment, and creativity as a triptych rather than isolated challenges, this talk aims to envision AI systems that are not only more capable and reliable, but also generative in expanding scientific discovery and human knowledge.