No to Generic Names in Prompts

Lately, I have been working on extracting information from different sources. I combined those sources into a prompt and asked the LLM to extract key details, justify them, and cite the source.

Everything worked well, but I used generic labels for different sources, like “Section 1” and “Section 2”. Those labels eventually leaked into some of the justification text. For example: “According to Section 2, this customer is eligible for a refund,” instead of “According to the Billing Policy, this customer is eligible for a refund.”

Internally, everyone understood those names. For end users, though, they were obviously unclear.

From now on, I will explicitly name sources instead of using generic labels. It makes the output clearer for everyone involved.