Cognitive Load Theory ↔ the context window — bounded workspaces & chunking

The conceptual spine of the Advanced track. Human cognition falters under overload
because working memory is sharply limited — the central claim of Cognitive Load
Theory — and modern LLMs exhibit a structurally parallel failure: they operate with
a bounded active workspace at inference time, allocating finite attention across a
fixed context window, so that as the sequence grows, earlier information receives
progressively less focus. The research literature frames these as shared mechanisms
(bounded workspaces, chunking) rather than loose metaphor. Teaching the parallel
gives students a durable mental model: the context window is the model's working
memory, and it can be overloaded exactly as a human's can.