Structured data exercise

Study Notes: Prompt Engineering with Stop Sequences and Message Prefilling

• Purpose of Techniques:
• Message Prefilling: Used to guide the Large Language Model (LLM) into a specific response format or content structure. It is not limited to just characters; it can include full instructions.
• Stop Sequences: Used to tell the LLM exactly when to cease generating output.

• Core Strategy for Clean Output:
• To prevent initial commentary, use a prefilled assistant message that starts with the desired formatting (e. g. , three backticks `).
• Set the stop sequence to the same formatting (e. g. , three backticks `) to ensure the LLM stops generating text once the code block is complete.

• Addressing Formatting Issues:
• Language Identifiers: When using code blocks (e. g. , three backticks), the LLM may automatically insert a language identifier (like bash) for syntax highlighting.
• Guiding Syntax: To prevent unwanted identifiers, explicitly include the desired language in the prefilled message (e. g. , bash `).
• Enforcing Output Structure: To prevent the LLM from adding extra commentary or generating multiple separate code blocks, use a highly specific prefilled instruction (e. g. , "Here are all three commands in a single block without any comments. ").

Takeaways

• Message prefilling guides the LLM into a specific response format or content structure, while stop sequences dictate exactly when the LLM must cease output generation.
• To achieve clean output, use a prefilled message that starts with the desired formatting and set the stop sequence to match that same formatting.
• To prevent the LLM from automatically inserting unwanted language identifiers in code blocks, explicitly include the desired language in the prefilled message.
• Highly specific prefilled instructions are required to enforce strict output structure and prevent the LLM from adding extra commentary or generating multiple blocks.