Prompt Patterns For PDS4 Information Modeling

Prompt Patterns provide general and reusable solutions to commonly occurring problems within specific con-texts while interacting with a Large Language Model (LLM) such as GPT4. A catalog of generic prompt engineering patterns [1] has been published to help users improve LLM results and to promote further re-search in prompt engineering. Software engineering software patterns are an analog to prompt patterns, providing reusable solutions to common problems in a particular context. The catalog defines five categories of prompt pat-terns including Input Semantics, Output Customization, Error Identification, Prompt Improvement, Inter-action, and Context Control. Prompt patterns from two of categories, Input Semantics and Output Customiza-tion have been applied to problems found in information modeling for the Planetary Data System (PDS). These patterns focus on defining input to and output from the LLM. The Input Semantics category deals with how an LLM understands the input and how it translates the input into something it can use to generate output. The Output Customization category focuses on constraining or tailoring the types, formats, or structure of the output generated by the LLM.