This folder holds golden prompt/response pairs for Google Duet AI training.
A golden prompt/response pair contains two parts:
- An example prompt/question to ask an LLM
- An ideal answer we would expect the LLM to generate
Each prompt/response pair is a markdown file with the following structure:
Prompt:
<markdown for prompt>
Response:
<markdown for response>
This folder includes the following types of golden prompts:
- Documentation lookup prompts
- Code generation prompts
- Code explanation prompts
Features of a good response:
- Uses official product names in the response (“Speech to text” → “Speech-to-Text”).
- Answers the question (correctly) with hyperlinks to the documentation.
- Includes a link to the corresponding source code.
- Includes a link to samples, if available.
Features of a good response:
- Starts with a brief introduction that explains the code sample.
- Includes information about how to find the reference documentation.
- Includes a link to the list of code samples.
- Provides well-documented code. Consider including an example of what the execution result looks like.
- Follows up with the user to ensure they don’t continue needlessly with false responses.
Features of a good response:
- Starts with a short overall description that tries to answer the question in the prompt.
- Grounds the algorithm in any well-known context, if appropriate. For example, this is an implementation of X, a well-known algorithm to do Y.
- Discusses the variables in the snippet and their purpose relative to the runtime.
- Discusses runtime and memory storage complexity.
- Notes any interesting features of the code, or opportunities for improvement (optimizations, refactoring, syntax best practices, etc.)
Folder structure:
learning/prompts/
├── code-explanation
│ ├── 01_io_kafka.md
│ └── ...
├── code-generation
│ ├── 01_io_kafka.md
│ └── ...
├── documentation-lookup
│ ├── 01_basic_learning_apache_beam.md
│ └── ...
└── README.md