The problem of context has in large part defined my career. It was the topic of one of my earliest publications - 'Why Equi Fails' - and of one of the very last reports I wrote at NRC. Context is why (Gary Marcus notwithstanding) rule-based systems fail. Context destroys ontologies - it's why it's ok to call something a 'tiger' one moment and absurd the very next (as Calvin and Hobbes both knew). This article is about context in the context of AI agents, and dives into the weeds, but it's worth the effort, if you look beyond the specific interactions depicted and think about this type of problem generally.
Today: Total: [Direct link] [Share]
