Fig. 6: General structure of prompt.
From: Detecting stigmatizing language in clinical notes with large language models for addiction care
When applying this prompt to the RAG approach, there are n context entries. When applying the in-context approach, all available context entries are provided. Finally, when applying to the Zero-Shot or SFT approach, there is no context subfield in the user portion of the prompt. System prompts prime the LLM to answer user prompts following certain directions. Hence the system prompt gives the LLM a scenario and an output format.
