RAG Interview Questions #20 - The Feedback Loop Trap

Why self-taught retrievers quietly destroy their own retrieval distribution while your eval metrics stay green, and the elite engineering shift from tracking accuracy to monitoring retrieval entropy.

You’re in a Staff ML Engineer interview at Google and the interviewer asks:

“Your REALM-style retriever learns from its own retrieval signal, no human labels. What breaks after six months in production?”

Don’t say: “Nothing, it’s self-supervised, it keeps improving as it sees more data.”

Wrong answer. You just described the failure, not the safeguard.

Here’…