Blog Posts

By: Dr. Robin Westerik

As Artificial Intelligence technologies rapidly make their way into our everyday lives, many of us grapple with making effective use of the AI driven tools and applications at our disposal.  One of the new gizmos we have are Large Language Models (LLMs) such as ChatGPT, Bard, and Bing Chat.  More and more, internet searches include LLM results.

LLMs provide us the ability to ask complex questions and quickly get good, quality answers. However, the results are only as good as the question—the old “garbage in, garbage out analogy” from statistical analysis. 

Composing effective queries for LLMs requires that users account for the model's capabilities and limitations.  To get the best results from our queries, we can follow some general guidelines and best practices.  In this blog, we cover 5 recommended practices for getting the results you want from AI queries.

Ask Actionable Questions/Commands

LLMs can access a large amount of information, distill it down, and present it to us in a matter of seconds.   However, that information may or may not be packaged in a manner relevant to our purpose.  To correct this, we need to ask actionable questions.  Actionable questions are questions that lead to specific actions or results.  In other words, give the LLM a specific task, such as “Explain long division” or “What is the most lucrative college major?”  By giving the LLM a specific framework for a response, you more effectively shape the results you see.

Be Specific

Along with actionable requests, one of the most critical factors in writing effective queries is to be specific. When you ask an open-ended or vague question, LLMs may struggle to provide a meaningful response. Instead, try to narrow down your query by providing more context, using specific keywords, and defining any ambiguous terms. For instance, if you're asking about the benefits of regular yoga practice, it is helpful to specify things like demographic information or whether you are interested in mental health or physical health benefits from your yoga.  For example, the results for “what are the benefits of yoga” may not pertain to you in terms of age, familiarity with yoga, or desired outcomes.  Adding specificity to your queries will improve your results.

Use Natural Language

LLMs are trained on vast amounts of text data, so it's essential to communicate with them in a way they understand – using natural language. Avoid jargon and technical terms unless absolutely necessary, as these may confuse the model. Instead, opt for clear, concise sentences that mimic how humans would ask questions. For example, instead of asking "What is the average correlation coefficient between stock prices and GDP growth rates?", rephrase it to "How does GDP affect the stock market?"

Be Concise

LLMs are designed to process large volumes of text quickly, so keep your queries brief and to the point. Too many words or phrases can overwhelm the model. Additionally, avoid asking multiple questions in one query.  Instead, ask separate questions that can be answered independently.  In this way, your queries are more likely to result in responses that include the specific information you need. 

Test and Refine

Finally, don't assume that your initial query will yield perfect results. LLMs are not infallible, and it's essential to test and refine your queries based on the responses you receive. Pay attention to any ambiguities or inaccuracies in the answers and adjust your query accordingly. Through this iterative process, you'll develop a deeper understanding of what works best for each specific model and improve the overall quality of your results.

Final Thoughts

These guidelines can help you craft effective queries and get worthwhile results when you use LLMs. With practice and patience, by asking actionable questions, being specific, using natural language, being concise, and testing and refining—and with a little practice and patience—you'll be using LLMs like a pro in no time.

Dr. Robin Westerik is the President of the Madison Education Group and holds a doctoral degree in business.  Dr. Westerik has been an educator for over 20 years and specializes in administration and operational effectiveness.