AI: Scaling Flaw

In this episode of The Capital Cycle Podcast, Edward Chancellor interviews Kai Chen, an Emerging Analyst at Marathon Asset Management, about artificial intelligence. Kai discusses the market's excitement over AI, driven by the scaling law, which suggests that increased computational power leads to better models. However, he points out the limitations of large language models, which excel at pattern matching but struggle with novel problems, as demonstrated by the ARC prize challenge. Kai also addresses the rising costs of AI development, the prisoner's dilemma among big tech companies, and the potential commoditization of AI services, cautioning that excessive investment in AI infrastructure, fueled by the assumption that the scaling law will continue, may lead to overcapacity and poor returns.