18706_Authority_Oct

6 Navigating AI's Limitations: The Challenge of Hallucinations Amidst the rapid generation of information, a phenomenon known as "hallucination" can occur with AI. While seemingly factual, AI models can, at times, generate responses that are entirely fabricated or inaccurate, despite sounding plausible. AI is fundamentally a pattern-matching and data-consolidation program; it does not possess consciousness or intent. Its responses are based on the patterns and correlations within its training data, regardless of the factual accuracy of the source material. (For a deeper dive, consider, “I Think Therefore I Hallucinate: Minds, Machines, and the Art of Being Wrong” by Sebastian Barros). A practical example of this limitation emerged when an AI chat responded that it does not record conversation history. Upon rephrasing the prompt to "Would you please re-read our prior conversations relating to the topic? Compile a summary of each discussion," the AI was then able to process and summarize the preceding dialogue, demonstrating the importance of precise prompting. Developing Best Practices: An Evolving Landscape Best practices for AI utilization are continually evolving. As noted by Cade Metz and Karen Weise in "A.I. Is Getting More Powerful, but Its Hallucinations Are Getting Worse," the technology's rapid advancement necessitates ongoing vigilance and adaptation in its application. Despite these limitations, the strategic use of AI offers significant benefits. In my quest for municipal authority KPIs, for example, the cost-free access to AI (specifically Gemini and ChatGPT) allowed for a robust exploration. Crafting targeted prompts enabled the identification of KPIs best suited for my organization, determination of optimal reporting frequencies, and selection of effective communication methods for conveying this information to stakeholders. Putting AI to Task: Real-World Applications Beyond KPI identification, AI demonstrated capabilities for broader operational improvements. Sample prompts included: e ngagIng aI wIth targeted questIons can sIgnIfIcantly aId In formulatIng Ideas , explorIng complex topIcs from dIverse angles , and IdentIfyIng new avenues for InquIry .