When I first started using Claude, the powerful AI model developed by Anthropic, I was prompting it similarly to how I did with ChatGPT.
After reading this guide, I’ve discovered that prompting Claude requires a unique approach. There is also some crossover of some intermediate techniques from ChatGPT, such as few-shot prompting and “thinking step-by-step” prompts.
In this blog post, I’ll share my insights after testing the techniques discovered in the guide – to help you unlock the full potential of Claude and achieve exceptional results.
1. Placing Long Documents Before Instructions
When working with longer documents, it’s crucial to position your question or instruction at the end of the prompt. Claude gives more weight to the information provided at the end of the prompt, so placing important instructions there ensures that they are prioritized. Here’s an example:
<doc> [Insert long document here] </doc>
Summarize the main points of the document above in a concise paragraph.
2. Leveraging XML Tags for Structured Prompts
Claude models are trained to recognize and respond to specific XML tag structures. By using these tags, you can clearly separate instructions, context, examples, and input data. This helps Claude better understand your requirements and generate more accurate outputs.
E.g. using <text> and </text> to include your reference texts.
Examples:
<doc> [Insert long document here] </doc>
Summarize the main points of the document above in a concise paragraph.
<instruction>Summarize the key points from the provided text.</instruction>
<text>Climate change is one of the most pressing issues of our time. Rising global temperatures, caused by increased greenhouse gas emissions, are leading to more frequent and intense natural disasters, such as hurricanes, wildfires, and droughts. To mitigate the impacts of climate change, it is crucial that we reduce our carbon footprint by adopting renewable energy sources, promoting energy efficiency, and implementing sustainable practices in all aspects of our lives.</text>
3. Assigning a Role and Staying in Character
Like with ChatGPT, one of the most effective ways to get the best out of Claude is by assigning it a specific role. When prompting Claude, start by clearly defining its role, such as “You are a world-renowned data scientist” or “You are a Pulitzer Prize-winning journalist.” This helps Claude understand the context and adapt its responses accordingly. For example:
You are Dr. Sophia, a leading expert in renewable energy. Provide a concise explanation of the benefits of solar power for residential homes. Begin your response with “[Dr. Sophia]”.
4. Searching for Relevant Sentences
To help Claude locate and reference specific information within a long document, you can instruct it to begin its output by identifying the most relevant sentence to the question at hand. This technique improves Claude’s ability to provide accurate and context-specific responses given that it has a 200,000 context window. Here’s an example:
<doc> [Insert long document here] </doc>
<instruction> Question: What are the primary causes of deforestation in the Amazon rainforest? Identify the most relevant sentence in the document that answers this question, and begin your response with: “The most relevant sentence is: [sentence]” </instruction>
5. Prompt Chaining for Complex Tasks
When faced with a complex task that cannot be easily broken down into subtasks within a single prompt, you can use a technique called prompt chaining. This involves splitting the task into separate prompts, with the output of one prompt serving as the input for the next. Here’s an example:. Here’s an example:
<instruction>Generate a thesis statement for an essay on the benefits of regular exercise.</instruction>
Claude Response: Regular exercise offers numerous physical and mental health benefits, improving overall well-being and quality of life.
<instruction>Using the thesis statement from the previous prompt, create an outline for a five-paragraph essay.</instruction>
<thesis>Regular exercise offers numerous physical and mental health benefits, improving overall well-being and quality of life.</thesis>
6. Letting Claude Say “I Don’t Know”
To prevent hallucinations and ensure that Claude provides accurate information, it’s important to allow it to admit when it doesn’t know the answer to a question. By including a simple phrase in your prompt, you can encourage Claude to be upfront about its limitations. For instance:
Prompt: <instruction>Answer the following question only if you know the answer or can make a well-informed guess; otherwise, tell me you don’t know it.</instruction>
<question>What is the capital city of Wakanda?</question>
Claude Response: I don’t know the answer to that question. Wakanda is a fictional country from the Marvel Universe, and as such, it does not have a real-world capital city.
7. Encouraging Step-by-Step Thinking
Another prompt that also works well with ChatGPT. Instructing Claude to think step-by-step can significantly improve its problem-solving abilities. By adding the phrase “Think step by step” to your prompt, you encourage Claude to break down complex issues and provide well-reasoned answers. For example:
Prompt: <instruction> A car travels at a constant speed of 60 miles per hour for 3 hours. How far does the car travel? Think step by step.</instruction>
Prompt: <instruction>A store sells apples for $1 each and oranges for $0.50 each. If a customer buys 5 apples and 8 oranges, how much change will they receive from a $20 bill? Think through the problem step-by-step.</instruction>
8. Providing Examples for Few-Shot Prompting
Few-shot prompting is a powerful technique that involves providing Claude with a few examples of the desired output format. This helps Claude understand the structure and style you expect in its responses. The more relevant examples you provide, the better Claude can adapt to your specific needs. Here’s how you can incorporate examples in your prompt:
<instruction>Generate a short poem about the beauty of nature.</instruction>
<examples>Example 1: Sunlight dances on the leaves, A gentle breeze whispers through the trees. Nature’s beauty, a sight to behold, A treasure more precious than gold. Example 2: In the meadow, flowers bloom, Their colors paint a vibrant room. The birds sing a joyful tune, Nature’s symphony, a perfect boon. </examples>
You may also like 📖
- 7 Time-Saving Tricks with ChatGPT
- World’s Largest Supercomputer for AI training is out – and available for you to use
- How ChatGPT Will Destabilize White-Collar Work
- ChatGPT’s Code Interpreter – Top 6 uses
- 7 Genius Hacks that every ChatGPT user should know
- 7 incredible ways that ChatGPT can help Python programmers
- 7 Key Traits of an Exceptional ChatGPT Prompt Engineer
- 35 ways to use the Code Interpreter of ChatGPT to make money
- 7 insanely complex Excel formulas that ChatGPT can help you with
- ChatGPT can See Images: 10 Best Ways to use the Multimodal Feature
- ChatGPT vs Bard: Example outputs compared, prompts, and our thoughts
- 6 super ways to use ChatGPT’s custom instructions feature
- Are ChatGPT’s Answers Unique?
- Study This: If You Want to Win ChatGPT Prompt Engineering Competitions
- How to Make ChatGPT Write Longer – 10 Ways (with prompts)
- How to Think Like an Expert ChatGPT Prompt Engineer: 7 Proven Strategies
- Generative AI versus Predictive AI
9. Breaking Complex Tasks into Steps
When dealing with complex tasks, it’s essential to break them down into smaller, manageable steps. By providing Claude with a clear, step-by-step approach, you can significantly improve the quality and accuracy of its outputs. For example:
Prompt: <instruction>Create a social media strategy for a new e-commerce store as described. Follow the steps below</instruction>
<store description> About the store </store description>
<steps>Step 1: Identify the target audience and their preferred social media platforms. Step 2: Define the brand’s unique selling proposition and key messaging. Step 3: Develop a content calendar with a mix of promotional, educational, and engaging posts. Step 4: Set specific, measurable, achievable, relevant, and time-bound (SMART) goals for the campaign. Step 5: Outline a budget for paid social media advertising and influencer partnerships.</steps>
10. Be Direct, Concise, and as Specific as Possible
Being clear and precise in your instructions is crucial for getting the best results from any large language model. It’s important to tell the model exactly what you want it to do using affirmative commands. This approach is more effective than stating what it should not do.
When you provide Claude with a prompt, including detailed context is essential, and specifying the correct XML tag for organizing this information can greatly enhance the model’s performance.
<instruction>Summarize the main ideas from the following article within the <text> tags.</instruction>
<text>{Insert article content here}</text>
