Language Models and Automated Reification

This article by Ben Recht, Leif Weatherby, and Tyler Shoemaker discusses the concept of reification—making abstract concepts into material objects—in the context of large language models (LLMs). It examines how commercial LLMs automate the creation of reality by skipping collective interpretation and argumentation, using examples like Chris Olah's appearance with Pope Leo and the historical g factor in psychometrics. The essay critiques mechanistic interpretability research in AI, drawing parallels to 20th-century psychometrics and warning against offloading causal reasoning to LLMs. It is a tech-focused analysis of AI's impact on knowledge creation and scientific practice.