A-RAG lets the AI choose how to search, what to read, and when to stop, instead of following a fixed recipe.
Humans keep a big-picture memory (a βmindscapeβ) when reading long things; this paper teaches AI to do the same.