PRODUCTION SCENARIO
A communications team asks a foundation model to quote a policy that was published after the model's training cutoff. The application sends no supporting documents with the prompt.
What is the main risk of relying on the model alone?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: It can state a policy detail confidently with no supporting source.
AWS lists presenting false information when it does not have the answer, and presenting out-of-date information when the user expects a current response, as known limitations of a model on its own. The text reads fluently either way, so the danger is a confident quote with nothing behind it. Refusing is possible but is not the main risk, because the model usually answers. Amazon Bedrock does not use one customer's prompts to serve another. Nothing slows down on document size here, because no document is sent with the prompt at all.
Official Source: https://aws.amazon.com/what-is/retrieval-augmented-generation/ ↗