Experience-Sensitive Game Learning: A Behavioral Study of Humans and Language Agents
The episode discusses a paper titled "Experience-Sensitive Game Learning: A Behavioral Study of Humans and Language Agents." The hosts analyze how humans show interpretable learning shifts from greedy to global strategies in games, contrasting this with language agents that show noisy, transient gains. They conclude that for AI agents to truly learn from experience, they need to convert reflections into executable decision procedures.