How Far Will Foundation Models Evolve?
Session Overview
Generative AI is evolving beyond simple chatbots into a general-purpose intelligence platform capable of understanding and executing human knowledge and actions. Yet today's foundation models still face significant limitations.
Challenges remain across long-horizon reasoning stability, real-world understanding, agent utilization, multimodal integration, data acquisition, and training costs — problems that must be solved before the next generation of models can move forward.
Drawing on firsthand experience building foundation models, this talk assesses where AI stands today and looks ahead to which technologies will define the competitive landscape over the next five years. It will also explore the opportunities Korea has in the global race to build foundation models.
Key Takeaways
- The evolution of generative AI from chatbots to general-purpose intelligence platforms
- Current limitations of foundation models: reasoning stability, real-world understanding, multimodal integration, data, and cost
- A five-year outlook on the technologies that will shape the foundation model race
- Korea's opportunity in the global foundation model competition
Speaker
Jeonghwan Lim holds a PhD in Mathematics from the University of Oxford. He began his career leading data analysis at PUBG before joining Samsung Research as a data scientist, where he worked on large-scale AI systems.
He currently serves as CEO of Motif Technologies, where he leads the development of foundation models. Drawing on his direct experience building these systems, his work focuses on the technical and strategic questions shaping the next generation of general-purpose AI.