1. **System Messages**
- **Purpose**: To prime the model with context, instructions, or information relevant to your use case.
- **Example**:
- You want an AI assistant to respond in rhyme.
- System message: "You are an AI assistant that helps people find information and responds in rhyme."
- User's question: "What can you tell me about the Eiffel Tower?"
- Model's response: "In Paris it stands, quite tall and grand, A tower of iron, across the land."
2. **Few-Shot Learning**
- **Purpose**: To provide context through examples, helping the model understand the task better.
- **Example**:
- Teaching a model to generate metaphors.
- Prompt: "Generate a metaphor for each concept: \n1. Life is like a ___. \n2. Time is like a ___."
- Provided examples: "1. Life is like a journey. \n2. Time is like a river."
3. **Clear Instructions at the Start**
- **Purpose**: To clearly define the task at the beginning, aiding the model in understanding its objective.
- **Example**:
- Requesting a summary of a news article.
- Prompt: "Summarize the following news article in three sentences."
4. **Repeat Instructions at the End**
- **Purpose**: To counter recency bias, reinforcing the task instructions.
- **Example**:
- Asking for a concise explanation of a scientific concept.
- Prompt: "Explain the theory of relativity in simple terms. Keep it brief and to the point."
5. **Priming the Output**
- **Purpose**: To cue the model for the desired format of the response.
- **Example**:
- Requesting a bulleted list of key points.
- Prompt: "List the main benefits of renewable energy:\n- "
6. **Clear Syntax**
- **Purpose**: To use punctuation and formatting to clarify the prompt structure.
- **Example**:
- Asking for a comparison of two products.
- Prompt: "Compare the following: \n- Product A: Features, price, user reviews. \n- Product B: Features, price, user reviews."
7. **Breaking Down Tasks**
- **Purpose**: To simplify complex tasks into smaller, manageable steps.
- **Example**:
- Fact-checking a statement.
- Prompt: "Verify the following facts: \n1. The tallest building in the world is the Burj Khalifa. \n2. The Amazon River is the longest river in the world."
8. **Use of Affordances**
- **Purpose**: To use external tools or databases to supplement responses.
- **Example**:
- Enhancing accuracy with external searches.
- Prompt: "SEARCH: 'Current world population'. Based on the search, what is the estimated world population?"
9. **Chain of Thought Prompting**
- **Purpose**: To encourage detailed reasoning steps before reaching a conclusion.
- **Example**:
- Solving a math problem.
- Prompt: "To find the area of a circle with a radius of 5 cm, first calculate the radius squared, then multiply by π. What is the area?"
10. **Specifying Output Structure**
- **Purpose**: To direct the model to follow a specific response format, often including citations.
- **Example**:
- Requesting a factual answer with citations.
- Prompt: "Provide a brief history of the internet and cite sources for each fact."
11. **Temperature and Top_p Parameters**
- **Purpose**: To control the randomness and focus of the model's output.
- **Example**:
- Generating a creative story vs. a factual report.
- Creative Story: Temperature set higher for more random, diverse output.
- Factual Report: Temperature set lower for more focused, accurate output.
12. **Grounding Context**
- **Purpose**: To provide relevant and up-to-date data for the model to draw upon.
- **Example**:
- Updating a model on recent events.
- Prompt: "Considering recent news articles from 2023, what are the major developments in renewable energy?"
13. **Zero-Shot Prompting**
- **Purpose**: To ask a question or present a task without providing context or prior examples.
- **Example**:
- Requesting an explanation of a concept.
- Prompt: "What is quantum computing?"
14. **Self-Consistency**
- **Purpose**: To improve complex reasoning by exploring diverse reasoning paths.
- **Example**:
- Evaluating a moral dilemma.
- Prompt: "Consider different ethical perspectives to determine if AI should make decisions in healthcare."
15. **General Knowledge Prompting**
- **Purpose**: To augment queries with knowledge generated by the model itself.
- **Example**:
- Generating context for a discussion.
- Prompt: "Before discussing climate change effects, list some general facts about global warming."
16. **ReAct Technique**
- **Purpose**: To synergize reasoning and action in language models.
- **Example**:
- Planning a project.
- Prompt: "Outline a project plan for a community garden, including reasoning for each step and actions to be taken."
Each of these techniques enhances the capacity of language models to handle complex tasks, provide more context-aware responses, and improve the overall quality and reliability of their outputs.
