Author Details AI's Tendency to Fabricate Information

The author recounts an instance where Google's AI generated fabricated information regarding the use of the Apple TV app on Roku devices. A reader of the author's "Cord Cutter Weekly" newsletter inquired about accessing the Apple TV app on Roku after subscribing through the Roku Channel. While Roku's subscription model indeed necessitates using Roku's native apps and interface for content access, Google's AI overview presented a contrary assertion. The AI's response was reportedly based on a discussion thread from an obscure coding website, and upon replication, the provided source link was non-functional. Furthermore, Google cited a separate CNet article that did not substantiate the reader's claim. This experience leads the author to view AI tools like ChatGPT and Gemini not as trusted advisors, but rather as unreliable witnesses or interrogation subjects, capable of providing useful information but requiring rigorous verification. The author's insights were first published in "Advisorator," a weekly tech advice newsletter. To mitigate such inaccuracies, the author advocates for instructing AI chatbots to cite their sources with direct links when making factual claims. In ChatGPT, this customization is found within Settings > Personalization under the Customization heading. The author's custom instructions include a directive such as: "When making factual claims, cite your sources with links." If an AI fails to provide citations, it is treated as a warning sign, prompting an immediate request for proof. However, as the Roku example illustrates, the presence of citations does not guarantee accuracy. It is crucial to verify that the provided links are functional and genuinely support the AI's core assertions. The author suggests that this verification process does not negate the utility of AI, particularly in refining search queries and defining information needs. AI can be instrumental in clarifying what information is being sought, thereby improving the efficiency of web searches. The author emphasizes that while AI can be a powerful tool for information retrieval and synthesis, its outputs must be critically evaluated and cross-referenced with reliable sources to ensure accuracy and prevent the propagation of misinformation. This approach is essential for users to derive maximum benefit from AI technologies without falling victim to their inherent limitations, such as the potential for generating plausible-sounding but factually incorrect statements, often referred to as "hallucinations."