When using AI systems, what are some best practices for ensuring the results you receive are accurate, relevant, and aligned with your original goals?

training it in such a way imrpoved its cognitive thinking

training it in such a way imrpoved its cognitive thinking

Keep the basics simple:

- Say exactly what you need. State the goal, audience, and decision the answer should support.

- Give the right context. Include the key constraints: timeline, country, internal rules, level of detail, and output format.

- Ask for a clear structure. For examples summarize in 5 points, compare 2 options, or flag assumptions.

- Check anything critical. Any sensitive, regulatory, numerical, or high-impact output should be verified against reliable sources and your internal controls.

- Review with a critical eye. Check that the answer is coherent, complete, not invented, and actually answers the question.

- Refine once or twice. If the answer is vague, narrow it: be more concrete, focus on operational risk, or use only reliable sources.

Simple example:

Instead of saying: Tell me the risks of AI, say:

I am preparing a note for a risk committee at a Canadian bank. Give me the 5 main risks of using an internal LLM, with one concrete example and one recommended control for each. Do not invent anything, and flag points that need validation.

In practice, the safest rule is: set the task clearly, then verify before acting.

If you’d like to continue the conversation or exchange views on AI risk in financial services, feel free to reach out to me on LinkedIn.

I'm continuing to focus on prompting, so I can get the best initial answer, then refine with feedback. I rarely take the first output.

Best practices to ensure AI results are accurate, relevant and aligned:

1. Use a structured framework like CREATE - Be specific with Context, Request, Examples, Adjustments, Type of output, and Extras. The clearer the prompt, the more aligned the result.

2. Iterative refinement & Human-in-the-Loop - Don't accept the first output. Review, validate against project data/sources, and use follow-up prompts to refine. We as PMs remain accountable.

3. Provide relevant context and constraints - Include project goals, stakeholder needs, and limitations so the AI doesn't hallucinate.

4. Fact-check and bias check - Cross-reference AI outputs with trusted sources, SMEs, and organizational data. Check for outdated info or bias.

5. Use advanced patterns when needed - ReAct for changing objectives, Chain-of-Thought for complex reasoning, and Flipped Interaction to let AI clarify unclear requirements.

Best practices to ensure AI results are accurate, relevant and aligned:

1. Use a structured framework like CREATE - Be specific with Context, Request, Examples, Adjustments, Type of output, and Extras. The clearer the prompt, the more aligned the result.

2. Iterative refinement & Human-in-the-Loop - Don't accept the first output. Review, validate against project data/sources, and use follow-up prompts to refine. We as PMs remain accountable.

3. Provide relevant context and constraints - Include project goals, stakeholder needs, and limitations so the AI doesn't hallucinate.

4. Fact-check and bias check - Cross-reference AI outputs with trusted sources, SMEs, and organizational data. Check for outdated info or bias.

5. Use advanced patterns when needed - ReAct for changing objectives, Chain-of-Thought for complex reasoning, and Flipped Interaction to let AI clarify unclear requirements.

Best practices to ensure AI results are accurate, relevant and aligned:

1. Use a structured framework like CREATE - Be specific with Context, Request, Examples, Adjustments, Type of output, and Extras. The clearer the prompt, the more aligned the result.

2. Iterative refinement & Human-in-the-Loop - Don't accept the first output. Review, validate against project data/sources, and use follow-up prompts to refine. We as PMs remain accountable.

3. Provide relevant context and constraints - Include project goals, stakeholder needs, and limitations so the AI doesn't hallucinate.

4. Fact-check and bias check - Cross-reference AI outputs with trusted sources, SMEs, and organizational data. Check for outdated info or bias.

5. Use advanced patterns when needed - ReAct for changing objectives, Chain-of-Thought for complex reasoning, and Flipped Interaction to let AI clarify unclear requirements.

Best practices to ensure AI results are accurate, relevant and aligned:

1. Use a structured framework like CREATE - Be specific with Context, Request, Examples, Adjustments, Type of output, and Extras. The clearer the prompt, the more aligned the result.

2. Iterative refinement & Human-in-the-Loop - Don't accept the first output. Review, validate against project data/sources, and use follow-up prompts to refine. We as PMs remain accountable.

3. Provide relevant context and constraints - Include project goals, stakeholder needs, and limitations so the AI doesn't hallucinate.

4. Fact-check and bias check - Cross-reference AI outputs with trusted sources, SMEs, and organizational data. Check for outdated info or bias.

5. Use advanced patterns when needed - ReAct for changing objectives, Chain-of-Thought for complex reasoning, and Flipped Interaction to let AI clarify unclear requirements.

Providing AI with clear project context by sharing key documents permitted by the organization, reviewing and validating its outputs, and iteratively guiding the conversation can help align the AI-generated contents with the project goals.