Sunghee Yun (He/Him)

Co-founder & CTO @ Erudio Bio, Inc., Chief Business Development Officer @ WeStory.ai, Advisor & Evangelist @ CryptoLab, Inc., KFAS-Salzburg Global Leadership Initiative Fellow @ Salzburg Global Seminar, Adjunct Professor @ Sogang University, Advisory Professor @ DGIST, Global Advisory Board Member @ Innovative Future Brain-Inspired Intelligence System Semiconductor of Sogang University, Network Expert @ Gerson Lehrman Group

Abstract

This presentation provides a comprehensive overview of Large Language Models (LLMs) and Generative AI, examining their technological foundations, industry applications, and market dynamics. The technical section explores the evolution from traditional RNN-based sequence-to-sequence models to the revolutionary Transformer architecture, highlighting how the “attention is all you need” mechanism solved fundamental limitations of recurrent models through parallelizable processing and the ability to handle arbitrarily long dependencies. The presentation details the mathematical foundations of attention mechanisms, including single-head scaled dot-product attention and multi-head attention, explaining how these simple yet powerful concepts enable LLMs like GPT, BERT, and Llama to achieve unprecedented performance in natural language processing tasks. Key architectural innovations such as self-attention, encoder-decoder attention, and multimodal learning capabilities are examined, along with their implications for diverse applications ranging from conversational AI to code generation.

The industry and market analysis reveals the explosive growth trajectory of AI technologies, with the global NLP market expected to reach $413.11B by 2032 and AI funding in Silicon Valley soaring to $17.9B in Q3 2023. The presentation discusses the competitive landscape dominated by big tech companies like OpenAI, Microsoft, Google, and Meta in foundation models, while highlighting opportunities for smaller players in niche markets and downstream applications. Significant attention is given to the semiconductor market transformation driven by AI demands, including the US CHIPS Act’s impact on global supply chains and the emergence of new competitors like AMD challenging NVIDIA’s dominance. The economic implications extend beyond technology, with the Federal Reserve actively studying AI’s potential impact on productivity, inflation, and employment across various sectors.

The final section addresses critical philosophical and practical questions surrounding AI development, including cognitive biases in both humans and machines, ethical considerations around AI consciousness and reasoning capabilities, and the ongoing debate about whether LLMs truly possess knowledge, belief, or reasoning abilities. Drawing on cognitive scientific perspectives, the presentation argues that while LLMs can convincingly mimic human-like responses, they fundamentally operate as sophisticated conditional probability models rather than systems with genuine understanding or consciousness. Recent developments like Kolmogorov-Arnold Networks (KAN) and Joint-Embedding Predictive Architecture (JEPA) are explored as potential paths forward, while Leopold Aschenbrenner’s predictions about AGI by 2027 and superintelligence thereafter frame discussions about AI’s future trajectory and the critical importance of alignment, safety, and responsible development practices.