AI Is Making a Lot of Big Promises, but It Can’t Even Properly Identify My Cat

Google's integration of its AI, Gemini, into Nest security cameras promised advanced smart home functionalities, including natural voice interactions, seamless device integration, and detailed video annotations. However, real-world testing of Gemini for Home revealed significant shortcomings, primarily in its accuracy and reliability. The author tested the AI in Google Nest Cam Indoor (Wired, 3rd Gen), Google Nest Cam Outdoor (Wired, 2nd Gen), and Google Nest Doorbell (Wired, 3rd Gen), noting a new level of intrusive monitoring coupled with escalating AI hallucinations. Initially, Gemini provided more detailed smartphone notifications compared to pre-Gemini Nest cameras, offering context like "Person walks into the room" instead of a simple "Person detected." With Familiar Face detection enabled, notifications became even more specific, identifying family members by name. The AI quickly escalated to providing a deluge of alerts, documenting mundane activities such as a person drinking water, a cat jumping on a couch, or items being tossed in the trash. Even a disguised person was accurately described as "A person wearing a cat mask and a black shirt walks into the room." Despite the initial detailed observations, Gemini soon began to generate significant inaccuracies. It misidentified a cat's actions, claiming it was scratching furniture when footage showed otherwise, and reported the presence of the author's husband when he was absent. The AI also frequently hallucinated, claiming multiple cats of different colors when only one orange cat was present, and insisted on the presence of a dog, chipmunk, hamster, and numerous mice, none of which existed. A daily Home Brief, meant to summarize events, often included these fabricated observations, leading to a state of constant misinformation. The article highlights that while some automations, such as triggering deck lights based on motion, worked reasonably well, the core AI powers of observation were deeply flawed. The Google Home app allows users to provide feedback on video clips via a thumbs-up/thumbs-down system, but with an average of 300 events per day, actively correcting the AI's mistakes proved impractical and time-consuming. This lack of user-friendly correction mechanisms exacerbates the problem of inaccurate AI interpretation. The author concludes that Gemini for Home, in its current iteration, undermines the fundamental purpose of security cameras, which is to provide unbiased and reliable observation. The constant mislabeling of people, objects, and animals, combined with AI hallucinations, renders the system untrustworthy and potentially dangerous. The article cites similar experiences from other users, further emphasizing the widespread nature of these issues. A Google spokesperson acknowledged that Gemini for Home is in early access and that the company is investing in improving accuracy, particularly regarding Familiar Faces identification. However, for now, the system is deemed unreliable for security purposes, as its flaws erode user trust and could lead to critical events being overlooked. #GoogleGemini #NestCameras #SmartHomeSecurity #AIHallucinations #HomeAutomation #TechReview #SmartHomeDevices #SecurityCameraFlaws #GoogleGemini #NestCameras #SmartHomeSecurity #AIHallucinations #HomeAutomation #TechReview #SmartHomeDevices #SecurityCameraFlaws