Hacker News Reader

I would say that disqualifies them from "understanding" anything. What they're doing is more like a broad search than pursuing a greater understanding

I agree - I view LLMs as semantic databases (of mostly stolen training data) with inference capabilities and NLG querying.

Semantics

The only convex hull that matters is the one that encloses the axioms.

Newsflash for you, humans are the same.