Director, Learning Design & Development| PMIAsheville, NC, United States
Validating and checking outputs is critical when working with AI systems like Generative AI. Such validation approaches may include establishing clear criteria, implementing strong testing protocols, and continuous refinement.
In your experience with AI, what are some best practices for ensuring the results you receive are accurate, relevant, and aligned with your original goals?
When you input questions or tasks to AI, be sure to use prompt engineering method with clear context and objectives, and always double check the outputs to ensure it is make sense and reasonable against your actual project data Saving Changes...
Kelly DobsonNorthside HospitalAtlanta, Ga, United States
One can approach LLMs as a pilot project where you can test with a small amount of information to ensure it is working properly then keep building on it. Also, what I have learned is that AI is only as smart as the information you give it, so users really need to focus on feeding it accurate and specific information/requests for it to work in an optimal manner and produce desired results. Saving Changes...
Many best practices have already been shared above which are relevant to address the requirements of accuracy, relevance and matching our goal requirements. Define the persona, clearly state the task/questions being asked supported with context by providing examples relevant to the situation. Of course, validate the response and iterate till we get desired answer in the format we need. CREATE methodology beautifully covers all these in a structured manner ! Saving Changes...
When using AI systems, some of the best practices for ensuring the results I receive are accurate, relevant, and aligned with my original goals is that utilize my prompts to the best of my abilities. The prompts that are utilized should be crisp and clear and task oriented. Also double check that they are closing the loop so that your receiving exactly what you asked for. A great example would be placing an order from a menu. So we keep things honest and pertinent. Saving Changes...
Here are my thoughts on ensuring AI results stay accurate, relevant, and goal-aligned: Start with Crystal-Clear Intent I've learned that vague requests produce vague results. Before engaging with AI, I take a moment to ask myself: "What exactly am I trying to achieve?" The more specific I am about my desired outcome, format, audience, and constraints, the better the AI can serve me. Treat It as a Conversation, Not a Command AI works best when I approach it iteratively. I don't expect perfection on the first try. Instead, I provide my initial request, review what comes back, then refine: "This is good, but can you make it more technical?" or "I need this shortened to fit one page." Each exchange gets me closer to what I actually need. Verify, Don't Just Accept This is crucial—I never blindly trust AI outputs, especially for facts, statistics, or technical details. I cross-reference important information, check sources when provided, and apply my own expertise and judgment. AI is a tool to augment my thinking, not replace it. Give Context Generously The AI doesn't know my background, my company's culture, or my project's history unless I share it. When I provide context. "I'm presenting this to executives who prefer data over anecdotes" or "Our team values collaborative language" the results align much better with my actual needs. Course-Correct Early and Often If I notice the AI veering off track, I interrupt immediately rather than letting it continue down the wrong path. A quick "Actually, let me clarify what I meant..." saves time and gets better results than trying to salvage misaligned output. Keep My Critical Thinking Engaged Perhaps most importantly, I stay actively involved in the process. I question whether the suggestions make sense, whether they align with my values and goals, and whether they serve my audience's needs. The AI is my collaborator, but I'm still the decision-maker. Saving Changes...
To ensure accuracy and alignment, provide clear, specific instructions with ample context. Use iterative prompting to refine outputs and strictly verify all factual claims. Employ techniques like "Chain-of-Thought" for complex reasoning. Treat the interaction as a dialogue, constantly guiding the AI back to your specific objectives to prevent drift.
Use a prompt formula such as CREATE to provide detailed guidance to the AI to ensure clear and specific results. Apply prompt engineering best practices, such as advanced prompt patterns or iterative prompt refinement, to align responses with project objectives. Ensure responses adjust for new data or information, especially in continuously changing projects, and request in-depth analysis or strategies to gain more actionable insights. Saving Changes...
A very good question and also difficult to answer as well. However you have to go to the basics and say as far as you are concerned, how well are you versed with the subject at hand ?. There are facts which the AI will generate and if you can verify these facts the more reliable the generated response will be. The fewer the facts then it means that the Generative AI response is far from meeting your original goals. Then it becomes very critical that you review the accuracy , relevancy and the alignment of the response to your original need. Unfortunately there are no clearly defined metrics that one can use a model to evaluate an AI generated response. So from my personal experience I basically restrict AI to an area where i have sound knowledge of , else it becomes almost impossible to verify details generated by an AI if you venture into unchartered territory. However with long usage and exposure your confidence also tend to increase as well. The best practice and protocol to follow would be to consult subject matter expects to validate the AI generated response before making critical decisions based on it to avoid any inherent associated risks which you might be not aware of.
Great point. When working in an area, where I have good knowledge, I use GenAI in discussion mode. Here I present the initial request with context and expected response format. But when using AI in an area I am unfamiliar, I ask the AI to provide sources for all provided information and insights, while refining my requests with each response. Saving Changes...
PMI Team Member
