Joon discusses the recent advancements in extending token limits for AI models, hinting at the potential for a million-token capacity. While this could enhance certain applications, he questions the necessity of such expansion for creating human-like agents, emphasizing the efficiency of decision-making in real-time scenarios.
