Google

Large Language Models can process text in various formats, not just standard prose. This allows for the use of structured notations like XML tags or control characters to represent complex data. It is also beneficial to explicitly define the desired structure for both the input and output within the prompt itself.

Using Structured Formats in Prompts

A practical method for structuring prompts is to use delimiters, such as quotation marks, to segregate different sections of the input text. To ensure the model correctly interprets this structure, the prompt should also include an explicit instruction explaining the role of the delimiters, for instance, by stating that 'the input text is delimited by double quotes.'

Specifying Input Structure Using Delimiters

Analyze the provided case study. Explain why the current approach is producing inconsistent results. Then, design a revised prompt that uses a structured format to guide the model toward generating a consistent, well-organized comparison. Justify the structural choices in your proposed prompt.

Improving Product Comparison Generation

A user wants to generate a recipe that can be easily processed by a computer program. Their current prompt is: 'Give me a recipe for chocolate chip cookies.' Rewrite this prompt to explicitly request the output in a structured format using distinct sections for elements like 'Title', 'Ingredients', and 'Instructions'.

Rewriting a Prompt for Structured Output

A developer is building a system to extract key information (product name, price, and stock status) from unstructured product descriptions. Their initial prompt, "Extract the product name, price, and stock status from the following text:", yields inconsistent and unreliable results. Which of the following revised prompts is most likely to produce consistently structured and accurate output?

Learn Before

Related