Chinese Room Argument
A person who doesn't understand any Chinese, given a sufficiently detailed rulebook, can mechanically swap Chinese symbols for other Chinese symbols according to the rules, and thereby make someone outside the room believe the person inside genuinely understands Chinese — this thought experiment argues that even if a system's output behaves as though it understands language, that doesn't prove genuine "understanding" actually happened inside the system. Half a century later, this is exactly the point where the argument lands most directly when applied to large language models.
intermediate
Orthogonality Thesis
An AI system's level of intelligence and the goals it pursues are, in principle, independent variables — a superintelligent system could just as easily pursue a trivial, bizarre, or catastrophic goal. Intelligence alone does not automatically lead toward "better" or more human-compatible values.
advanced