The practice of "theory building" is especially endangered by the current LLM uptake. Not because LLMs are good at it, because they are bad at it. Worse, they sometimes actively hinder theory building, because it usually rests on actually having deep experience with some techniques: 1: you do something over and over, 2: it becomes a second nature, you just get the feel for it, 3: you intellectualize this intuitive understanding, 4: you discover a new structure, or a better way to express/teach/learn/use an existing one.

Theory building was always kind of niche, or at least less prestigious (in comparison with solving well-known problems). Now, because the mad problem solvers are pulling on the blanket so much harder, the marginalization is getting worse.