.png)
"I don’t use ChatGPT because it's bad. I can't do anything with the answers it gives me." This is the feedback I’ve received from many different individuals after testing the Large Language Model (LLM) for the first time. Given that we are still at the forefront of the AI revolution, this is not an invalid response. In our the first episode of our Podcast, we delve into how ChatGPT 3.5 falters when compared to the free version of Google Gemini, and even Claude.
However, what I often find from inquiring further with these people is that while Chat GPT’s output was sub-par, so were their prompts.. At this stage in the development of generative AI, understanding how to properly prompt LLMs is critical to achieving your desired output. "Prompt engineering," or the process of providing a detailed prompt with clear expectations, may seem counterintuitive when dealing with ChatGPT and similar models that are supposed to do all the work. Yet, this process is key to preventing disinformation in your outputs and saves you the hassle of repeating your prompts or piecing together your desired outcome from various responses.
1. Clearly Define Your Goal
You'll discover that large language models can generate surprisingly specific solutions when they understand your goal. ChatGPT is trained on a vast dataset, enabling it to craft solutions that pertain to an immeasurable range of scenarios.
Tips:
Begin your prompt with an action verb such as "generate," "summarize," or "analyze."
Simple prompt example: "Generate a 4-week strength training regimen."
Complex prompt example: "Analyze the collected user feedback from my surveys, summarize the top 3 takeaways with a focus on improvement, and categorize the rest based on importance."
Consistently utilizing action verbs within your prompts helps Chat GPT stay focused on the tasks at hand when providing an output. Using action verbs clearly defines those tasks.
2. Contextualize Your Scenario
Context allows Chat GPT to craft outputs that provide solutions for every aspect of your described situation. The more specific you are, the more descriptive the feedback will be. You will find that Chat GPT provides shorter, simpler answers when it is prompted in an insufficient manner. Redundancy is better than inadequacy. Allowing the language model to understand as much as possible about your situation provides for outputs that do not neglect any aspect of your situation. When you want specific answers, you have to provide specific details.
Tip: Ensure that your context answers the following key questions:
What is the user’s background?
What does success look like?
What environment are they in?
Example prompt: "I’m a 250lb male aiming to lose 20lbs of fat in 3 months. Due to my schedule, I only have time to cook meals on Sunday afternoons. Please provide a three-month diet program to assist me."
The first sentence in this example describes the user's background, and clearly defines the goal to convey what success looks like. The second sentence describes their environment, while the third makes a request of the chatbot. Prompting Chat GPT with the third sentence alone would produce a diet program that neglects the health and time factors of this specific situation, and would produce an output that comes off as a general suggestion.
3. Provide Examples
Research shows that including a relevant example in your prompt dramatically improves the output quality. Examples allow ChatGPT to mimic the style, structure, and tone found within the example, deviating from the generic ChatGPT response style.
Example prompt: "Generate a job description using the format of this existing job description attached below. [paste example]"
Providing an examples offers guidelines for Chat GPT to utilize when crafting your response. You can think of it as a way to introduce new data into its trained dataset, giving it a specific perspective that it would not have otherwise.
4. Request a Persona
Suggesting personas enables ChatGPT to draw from its training data to mimic responses that a particular persona is likely to provide. Personas can be job positions, characters, or even celebrities.
Example prompts: "You are a physical therapist specializing in athlete recovery. Generate a 2-month program that helps me recover from my tennis elbow." "Rewrite this poem in the style of Eminem."
Personas offer a particular uniqueness to responses, and allows Chat GPT to infer on what is most likely to appropriately be said from an individual in that position. When you request that Chat GPT "think" as a physical therapist to provide solutions for a health scenario, it will tailor the response to be from the perspective of physical therapy, and not just a general solution. Simply asking the chatbot for help recovering from tennis elbow would be likely to generate a response that provides a program limited to over the counter medications and rest.
5. Provide a Format
Prompt formats refer to the layout and structure of the requested output. ChatGPT can generate content in various formats flawlessly:
Table format
Bullet points
Email format
Code blocks
Paragraphs
Markdown
Example prompt: "Analyze this data of my personal strength training regimen over the last 4 weeks and act as a personal trainer to organize the workouts in a table format with headers that provide a complete exercise program for each day of the week. Ensure that the muscle groups exercised alternate appropriately on different days. [paste recorded data]"
When requested, Chat GPT is very good at generating and organizing information in the formats listed. Otherwise, its default setting is to provide written responses in paragraphs.
6. Request a Tone
Using adjectives to request a specific tone allows ChatGPT to generate outputs that sound more "human." This clarifies your expectations in the prompt and creates a cohesive overall tone. A specific tone ensures your content meets your informational needs and resonates with your intended audience.
Example prompts:
"Generate your output in a conversational tone of voice."
"Please analyze this data in a formal tone of voice."
"Express enthusiasm in your response."
Following these steps significantly increases the likelihood of receiving quality outputs from ChatGPT and similar LLMs on the first try. The common denominator in all these steps is specificity. Providing as many details as possible about your goals and the conditions of your scenario allows the chatbot to craft a thorough response that meets all your expectations. For additional information on prompting, we offer a comprehensive course on the topic, along with other insightful courses on a broad range of AI related topics and skills to ensure you are properly leveraging the technology to achieve maximum efficiency for your business and personal life.
Comments