Retrieval-Augmented Generation
The architecture behind every connected answer engine: retrieve relevant fragments from a live index, place them in the context alongside the question, and require the model to answer from those fragments rather than from memory.
Analogy Mental model
A closed-book examination answered from a textbook memorised two years ago, against an open-book examination where the candidate may consult ten freshly printed pages and copy the correct figure from them.
In practice What you actually do
The acronym maps directly onto three areas of work. Retrieval depends on conventional ranking plus semantic relevance, so the page has to be selected. Augmentation depends on clean semantic markup and sensible paragraph boundaries, so the passage survives intact. Generation depends on density, specificity, and supporting evidence, so your sentence is the one adopted.
Where people get burned Pitfalls
Explaining this relationship clearly is the single strongest answer to the standard interview question about how search and generative visibility relate, because it shows the mechanisms are joined rather than opposed.
This entry was written in English from the Chinese original, not machine-translated. Read the Chinese version →