Cameron Yetman

The extraordinary success of recent Large Language Models (LLMs) on a diverse array of tasks has led to an explosion of scientific and philosophical theorizing aimed at explaining how they do what they do. Unfortunately, disagreement over fundamental theoretical issues has led to stalemate, with entrenched camps of LLM optimists and pessimists often committed to very different views of how these systems work. Overcoming stalemate requires agreement on fundamental questions, and the goal of this paper is to address one such question, namely: is LLM behavior driven partly by representation-based information processing of the sort implicated in biological cognition, or is it driven entirely by processes of memorization and stochastic table look-up? This is a question about what kind of algorithm LLMs implement, and the answer carries serious implications for higher level questions about whether these systems have beliefs, intentions, concepts, knowledge, and understanding. I argue that LLM behavior is partially driven by representation-based information processing, and then I describe and defend a series of practical techniques for investigating these representations and developing explanations on their basis. The resulting account provides a groundwork for future theorizing about language models and their successors.

The discipline of philosophy has been critiqued from both within and outside itself. One brand of external critique is associated with Conceptual Metaphor Theory (CMT), the view that human cognition is partially structured by pervasive and automatic mappings between conceptual domains. Most notably, Lakoff and Johnson (1999) claimed that many central philosophical concepts and arguments rely on an unacknowledged metaphorical substructure, and that this structure has sometimes led philosophy astray. The purpose of this paper is to argue that Lakoff and Johnson's critique is anticipated by the work of post‐Tractarian Wittgenstein and his student, Margaret MacDonald. In the Blue Book, Wittgenstein outlines a method for identifying and resolving philosophical puzzles generated by misused grammatical analogies, although his discussion lacks a precise characterization of exactly how and why such analogies lead to trouble. In a 1938 paper, MacDonald offers such a characterization, which I outline and then connect back to Wittgenstein. In addition to this interpretive work, I supplement Wittgenstein and MacDonald's diagnosis using evidence from CMT which suggests that linguistic metaphors and analogies often originate in or are motivated by more fundamental analogical mappings in cognition. The supplemented account carries implications for how philosophical arguments ought to be formulated and critiqued.

For A. J. Ayer, the occurrence of delusions confutes the notion that we perceive the world directly. He argues instead that perceptions are caused by immaterial “sense data” which somehow represent the properties of material things to us in our experiences. J. L. Austin systematically rejects Ayer’s claims, arguing that the occurrence of delusions does not preclude the possibility of direct perception, and that, indeed, our normal perception is direct. I challenge both philosophers’ ideas by examining how they deal with the phenomenon of colour.