The Ultimate Guide to AI Prompting: Tools, Tips, and Frameworks for Success
Prompt engineering influences the performance and utility of LLMs. The quality of input will always determine the relevance and accuracy of the AI’s response, which makes prompt engineering a crucial skill set for individuals and teams seeking to harness the full potential of these tools that allow us to unlock insights, solve problems and generate useful questions and ideas. Below, I’ve compiled the most useful tips and best practices to get you started.
I love ❤️ a good acronym! That’s what first caught my attention when I learned of Kyle Balmer’s RISEN® prompt framework. It basically goes like this:
R - Role: give your AI a role, telling them to ‘act as a certain type of person’ (e.g., act as a professional copywriter or an SEO optimizer who prepares blog articles)
I - Instructions: provide instructions on what it is you want it to produce. For example instruct AI to write a blog article about ___.
S - Steps: Detail the required steps with as much accuracy as possible. E.g., Start with a really interesting hook to grab the attention of the reader, give me three main body points (each with examples and finish on a strong call to action which will get people to subscribe to a newsletter)
E - End Goal: Tell ChatGPT exactly what we expect to see (e.g., this blog article is aimed at a specific audience of taolu practitioners); give the EXACT ANGLE in the end goal prompt
N - Narrowing: Give constraints (e.g., make the blog article between 500-750 words, do not use overly flowery language, make sure to use human sounding language). Give constraints and narrow the scope of what the AI will produce — save the prompt in Notion or a spreadsheet, so that you can reuse it every time you want to write a blog article. Build a library of prompts for your business.
So what do the experts at large say about how to prompt like a pro?
Cautiously word your prompts. Just one or two words being edited, omitted or changed can drastically make or fail how AI interprets your request. Be mindful, thorough, and complete when finalizing your prompts. If working on a team, get inputs from colleagues and develop a set of guidelines that are updated and optimized frequently to ensure accuracy and relevancy given this ever-changing technology.
Use at least one example or reference in your prompt to help ensure desired results. If you provide an LLM with just one example in a prompt (giving context to your intended output), this significantly improves how well the AI responds to the prompt. Often, one example will do, as proven by the one-shot or single-shot approach.
It’s best to test this approach as you build your prompting library and your AI prompting techniques. There will be times when providing more than one example is key.
Establish a timeframe. Set time parameters around the research to ensure your prompt returns insights on the most up-to-date and relevant information (e.g., find recent articles about ___ from the last 30 days).
Use dynamic variables. This allows you to reference existing data in your prompt, like account or contact names, so you can research with specific data points (e.g., find recent news about the company '{{account.name}}' with the website '{{account.website_url}}' from the last 3 months.)
Get jiggy with the “Three Experts” Prompt: It borrows some ideas from popular prompting techniques like chain-of-thought and tree-of-thought, in a simple and straightforward way. Instead of getting a single answer result, you get multiple perspectives debating the merits of different approaches.This yields results for just about any scenario, and is pretty easy to put into practice.
Example:
Can you act as 3 different experts having a discussion about [your topic]?
Each should have different backgrounds in [relevant fields].
Have them discuss the pros and cons of [specific question/problem], considering different approaches and practical implications.
Make sure they engage with each other's points and work toward a practical conclusion.
Consider potential biases and ethical implications. Avoid biases if you want fair results. AI models pick up information from a ton of data from multiple sources like websites and articles. However, sometimes, this data might have some biases or unfair opinions (just like irl, where stereotypes and misinformation are all too common). Due to this, the AI could end up producing biased or inaccurate responses by mistake. Use words that are inclusive, impartial, and free of bias. For example, instead of saying "suffers from a disability," say "a person with a disability." Don't write prompts that make assumptions about people based on gender, race, or age. For example, instead of saying, "Older people struggle using technology", say, "Some individuals might have a hard time with new technology, no matter how old they are."
Consider employing the Socratic method to your AI prompts. Benefits of this method include: improved critical thinking, deeper understanding and self discovery via uncovering new perspectives and insights. This approach involves asking probing questions to challenge assumptions and encourage deeper exploration of the topic, fostering critical thinking and self-reflection. Steps include:
State your opinion or idea
ChatGPT responds with a series of open ended questions designed to challenge your assumptions
Reflect and refine: consider each question carefully and respond with thoughtful answers, improving upon your initial understanding of the subject being discussed
Remember, there are several types of AI prompts. Which one you use depends on what you want to achieve.
Creative: Create unique content such as text, images, audio, and videos, often stylized to meet specific instructions, e.g. “Write in the style of Charles Dickens.”
Instructional: Offer guidance, such as customer service. An instructional prompt could be: “Explain the steps a customer needs to take to upgrade their account.”
Informational: Gather and synthesize information, either from training data or uploaded resources such as PDF files. Use informational AI prompts to get summaries or insights.
Listicle: Compile lists, from simple to more detailed and structured ones. For instance, ask the AI to generate a list of blog titles or theme ideas.
Interactive: Mimic real-life conversations, such as mock interviews or role-playing scenarios.
Reasoning: Analyze information and draw conclusions. For example, use a prompt to identify a target market from sales data.
Keyword: Point your AI tool in the right direction, whether pulling insights from data or creating images by using specific keywords.
If this seems overwhelming and you’re looking to generate prompts that get the job done quickly and efficiently, I strongly recommend employing the Anthropic Console (console.anthropic.com) - it allows users to provide a basic description of the task they want an AI model (like Claude or ChatGPT) to perform, and the API will then automatically generate a well-structured, detailed prompt that can be used to get the best possible results from the model. It's a tool that helps users create effective prompts without needing in-depth prompt engineering knowledge by providing a starting point based on their desired outcome.
I certainly hope you found these AI prompting insights helpful and useful! For more AI tools and industry insights get in touch with me here at Dulcet Partners.
