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?
Some of the best practices include being specific, providing examples, providing AI with company specifics like assumptions, constraints, regulations that are specific to the company, otherwise the AI may respond with generics that are less useful. Saving Changes...
It is critical to formulate your prompts to describe clearly the context and ask clear questions. When the problem or question is complex it’s good to use prompt chaining where you split the topic into chained questions where you can give and ask feedback at each step making the responses more relevant to the specific scenario you are targeting. Through an iterative process both the questions asked and responses can better converge to provide the desired answers that are suited to the context and environment or application. Saving Changes...
With over 25 years’ experience towards familiarizing myself in the IT domain with PMI certification, I assist in ensuring that the results of AI seamlessly fit within the overall AI objectives. In order to keep the role of AI focused on what it hopes to achieve, the problem space needs to be well defined and we need to be setting clear measurable outcomes right from the outset.
It’s all about continuous monitoring and validation. Early detection of discrepancies in the output can only be gained by regularly reviewing AI outputs against the set goals. In addition, feedback from cross functional teams and domain experts further improves the quality of the results with their feedback giving the AI additional insights in its processing. With AI and human expertise combined, we’re able to keep precision and relevance through the project lifecycle. Saving Changes...
Tracey CreweHR Project Manager| University of WaterlooWaterloo, Ontario, Canada
As with any new tool being implemented, consider using it in parallel with existing tools & techniques. Be specific, and continuously validate responses received. Saving Changes...
In my experience providing detailed prompts, examples, and also reference sources the more accurate the result. I have also validated AI's responses by checking is sources for information. Saving Changes...
To ensure accurate, relevant, and goal-aligned results from AI systems, I would use the RTF (Role, Task, Format) and CREATE (Character, Request, Example, Adjustment, Type, Evaluation) frameworks to craft prompts effectively. I’d start by defining my RTF—considering the role the AI is to take, the specific task it should perform, and the desired format of its output. Then, I’d apply CREATE to further refine: shaping the AI’s persona (Character), clearly stating my question or request, providing an Example if relevant, adjusting language or specificity as needed, choosing the Type of output, and finally, Evaluating the response. I would repeat this process, refining my prompts based on each response, to progressively enhance alignment with my goals. Saving Changes...
ensuring that AI-generated results are accurate, relevant, and aligned with your goals requires a blend of clear communication, iterative refinement, and careful evaluation. Here are some best practices:
1. Define Clear Objectives
Clearly outline the purpose of your AI usage and the specific outcomes you want. The more precise your goals, the better the AI can align with them.
2. Ask Specific and Contextualised Questions
Avoid vague prompts; instead, provide specific details, context, and examples. For example, if you're asking for design advice, mention the style, audience, and medium you have in mind.
3. Use Iterative Refinement
Begin with a general prompt and refine it based on the AI's responses. This iterative approach helps you clarify and adjust your request for more targeted outputs.
4. Provide Relevant Background Information
If the task requires context (e.g., industry-specific terms or a particular knowledge area), make sure the AI has all relevant details to avoid general or inaccurate responses.
5. Cross-Check Critical Information
For high-stakes or technical information, validate the AI’s output against trusted sources or seek expert opinions to confirm accuracy.
6. Leverage Follow-Up Questions for Clarity
If an answer is ambiguous or lacks depth, follow up with clarifying questions. Asking for explanations or examples can also provide a deeper understanding of the response.
7. Request Alternative Perspectives
Sometimes, asking the AI to offer multiple viewpoints or rephrase answers can reveal additional insights or solutions you might not have considered initially.
8. Review and Apply Domain-Specific Knowledge
Ensure that responses align with the nuances of your field. If you have specific standards or practices, communicate these to the AI for more tailored guidance.
9. Utilize AI Tools in Combination with Human Judgement
AI can generate ideas and analyse data, but applying critical thinking and domain expertise is essential for final decision-making.
10. Give Feedback on Responses
When possible, indicate what worked and what didn’t to fine-tune interactions and improve the relevance and quality of future outputs. Saving Changes...
Anonymous
Ensure you are using the appropriate model, RTF vs CREATE based on your needs. Also ensure you check the output to make sure it includes the level of detail desired Saving Changes...
PMI Team Member
