This chapter focuses on a classic study of human problem solving, Luchins, A. S. (1942). Mechanization in problem solving: The effect of einstellung. Can a predictive model solve the problems Luchins used? Does it show similar effects of experience to those Luchins showed? The chapter also argues that LLMs demonstrate natural semantics, in which meaning is determined by the context of use, rather than artificial semantics, in which meaning is determined by rules of interpretation. This distinction reflects arguments made by sociologist Harold Garfinkel.