Have you experienced the frustration of repeatedly asking Chat GPT or Bard or MidJourney (etc.) a relatively simple thing, and not getting the results you were looking for? You are not alone! As Generative Artificial Intelligence (AI), particularly language models like ChatGPT, have started to carve out a niche and offer immense potential, they come with their unique challenges, one of which is how to work efficiently with them.
Prompt engineering is the art and science of crafting effective inputs to guide an AI model towards generating the desired output. In less fancy terms: prompts are our inputs, and the better we get at giving great prompts, the better the output will be. Although plugins like Prompt Perfect can help you, they still won’t do your job, so don’t rely solely on them.
Using ChatGPT and similar tools indeed has its risks: it may generate its answer from hallucinated sources or fail to provide sources altogether, and in most cases, the onus falls upon us, users, to verify the information - just to name a few. Despite its limitations, it is quickly becoming a must-have ally, which brings us to the crux of the matter - how do we ask the right questions?
Whether you're a seasoned user of AI or just starting to dip your toes into this brave new world, these guidelines should make your journey smoother and more rewarding.
The more specific your prompt, the more accurate and focused the response is likely to be.
Less specific: “Tell me about heart disease.”
More specific: “What are the most common risk factors for coronary artery disease?”
Describe exactly what kind of output you are looking for. It also helps highlight what’s important for you about the output.
“I’d like to get a short list of 5 ideas for a Youtube video on the future of healthcare AI.”
Consider the discussion you are having with ChatGPT as a discussion you would have with a person you just met who might still be able to answer your questions and address your challenges. The better you describe your goal, your target audience, and the tone you would like to use, the more fitting the output will be.
“I’m writing an article about tips and tricks for ChatGPT prompt engineering for people working in healthcare. Can you please list a few of those tips and tricks with some specific prompt examples? My audience consists of medical professionals who are not AI experts. Please use a friendly, approachable tone.”
The style of your prompt can significantly impact the answer. Try different formats like asking to generate a list, provide a summary, or explain like I’m five (ELI5), etc.
Direct Question: “What are the symptoms of COVID-19?”
Request for List: “List all the potential symptoms of COVID-19.”
Request for Summary: “Summarize the key symptoms and progression of COVID-19.”
ELI5: “Explain the symptoms of COVID-19 like I’m five.”
This way, you can get the desired process of getting the information or input you were looking for.
“Act as a Data Scientist and explain Prompt Engineering to a physician.”
“Act as my nutritionist and give me tips about a balanced Mediterranean diet.”
Unless you are simply amazing at prompt engineering, you rarely get the best response you were looking for after the first prompt. So feel free to refine your question. You are also encouraged to ask it to modify the output based on its previous response. If you’re not getting the output you want, try guiding the model with continuation prompts. This can be more effective than trying to get the perfect response in a single prompt. You can also make your prompt a conversation: ask the AI to think step by step or to debate the pros and cons before settling on an answer.
You: “What causes diabetes?”
AI: “Diabetes is primarily caused by the body’s inability to produce or effectively use insulin…”
You (refining the question): “What factors lead to the body’s inability to produce insulin?”
AI: “The body’s inability to produce insulin can be caused by autoimmune destruction of the beta cells in the pancreas, as in Type 1 diabetes…”
You can easily go back to a specific discussion by clicking on the right thread in the left column. This way, you don’t have to start all over again but can just continue a discussion you already had with ChatGPT.
Open-ended questions often yield more comprehensive responses.
Closed question: “Is exercise important for patients with osteoporosis?”
Open question: “How does regular physical activity benefit patients with osteoporosis?”
If there is something you are not satisfied with or don’t understand based on its response, first of all, tell ChatGPT that you don’t understand the answer and ask it to provide an example.
As you navigate the world of AI interactions, keep these final tips in mind to achieve the best results:
Avoid confusing the AI by using clear and straightforward prompts. Provide the necessary information without unnecessary complexity or ambiguity, allowing the AI model to understand your intended task easily.
Ensure that your prompts incorporate relevant contextual cues. By providing the right context, you enable the AI model to deliver more accurate and relevant responses. Be specific to the task at hand, avoiding generic or vague instructions.
Adopt anatural and conversational tone in your prompts. Imagine you're interacting with a knowledgeable person and use accessible language. This approach fosters a more human-like interaction and enhances your understanding of the AI's output.
By following these guidelines, you'll be well-equipped to craft effective prompts that optimize your AI interactions. Enjoy exploring the limitless possibilities of AI and uncovering valuable insights along the way. Happy prompting! 🚀✨