Understanding Text Delimiters

• Delimiters are characters or sequences separating or enclosing pieces of text.
• They aid AI models in distinguishing different parts of the input.
• Effective delimiters enhance the model's ability to understand and process instructions, leading to more accurate and relevant responses.

Common Delimiters

• Quotation marks (", ', `) are frequently used for enclosing text or code.
• Backticks (`) are helpful for delineating code snippets within a larger text.
• Angle brackets (<>) can contain specific data or instructions.
• Dashes (--) or underscores (__) can separate sections or highlight key phrases.
• XML-style tags (e.g., <tag> and </tag>) are suitable for structured data.
• Colons (:) can indicate a label or category.
• Periods (.) can separate sentences.
• Commas (,) can separate items in a list.
• Special characters like #, $, %, and * can denote specific instructions or formatting.

Effective Use of Delimiters

• Delimiters should clearly separate instructions from input data.
• Maintain consistency by using the same delimiters throughout the prompt.
• When nesting delimiters, ensure each level is distinct.
• Select delimiters unlikely to appear within the content being delimited to avoid ambiguity.
• Choose delimiters that match the context and the information type processed.

Examples of Effective Prompts with Delimiters

• "Translate the following English text into French: Hello, how are you?" uses backticks to enclose the text for translation.
• "Summarize the article within the <article> and </article> tags" leverages XML-style tags to define the article.
• "Extract all names from the following list: Names: John, Jane, Michael, Emily" employs a colon to separate the "Names" label from the list.
• "Analyze the code below:\n\ndef hello():\n print('Hello, world!')\n" uses triple backticks for a code block.
• "Classify these sentences:\n- Sentence 1: The cat is black.\n- Sentence 2: The sky is blue" uses dashes to separate sentences.
• "Identify the key entities in this passage: <passage>The CEO of Apple, Tim Cook, announced new products yesterday.</passage>" uses angle brackets and a descriptive tag.

Pitfalls to Avoid

• Avoid mixing different types of delimiters without a clear reason, as this creates inconsistent delimiters.
• Characters appearing within the content itself can cause ambiguous delimiters, leading to confusion.
• Nesting delimiters of the same type without proper escape characters leads to overlapping delimiters.
• Failing to properly close delimiters results in missing delimiters, causing incorrect parsing.
• Using delimiters when unnecessary can add noise to the prompt from unnecessary delimiters.

Best Practices

• Start with basic delimiters like quotes or backticks before using more complex options.
• Evaluate how the model responds to different delimiters and choose those yielding the best results through thorough testing.
• Provide clear examples in the prompt when using custom delimiters.
• Document the purpose and usage of each delimiter when creating a complex prompt.
• Adjust delimiters continuously based on the model's performance and specific task requirements through iteration and refinement.