Gemini has introduced Personal Intelligence, a new feature designed to make the AI assistant more helpful by understanding and using your personal data across apps like Gmail, Google Photos, and others—while keeping privacy at the core. Personal Intelligence excels at two key tasks: reasoning across multiple sources of information and retrieving specific details from emails, photos, or videos to answer questions accurately. It often combines these abilities to deliver tailored responses. For example, when preparing to buy new tires for a 2019 Honda minivan, Gemini didn’t just find the correct tire size—it suggested options based on driving habits and weather conditions, referencing past family trips to Oklahoma captured in Google Photos. It then pulled ratings and prices for each option. Later, when needing the license plate number at the dealership, Gemini retrieved it instantly from a photo in Photos and even identified the van’s trim level by searching through Gmail. Beyond practical tasks, Personal Intelligence helps with everyday decisions. It’s recommended travel plans for spring break that align with family interests, avoiding crowded tourist spots and instead suggesting an overnight train ride with board games suited to the group. It learns from your history—your emails, photos, and past trips—but only when you choose to connect your apps. Privacy is central to the design. App connections are off by default, and you control which apps to link and when to disconnect. Data stays within Google’s secure environment, so you don’t have to send private information to third parties. When Gemini uses your data, it clearly shows where the information came from, so you can verify it. If a response seems off, you can correct it directly—like saying, “Remember, I prefer window seats”—and the system learns from your feedback. You can also use temporary chats to avoid personalization or regenerate responses without any personal data. Gemini does not train directly on your inbox or photo library. Instead, it uses anonymized and filtered data from your interactions—like your prompts and the model’s responses—to improve over time. For instance, it learns how to find your license plate number when asked, but it doesn’t learn your actual plate number. The system includes safeguards for sensitive topics, avoiding assumptions about health or personal details unless you bring them up. It may still make mistakes, like over-personalizing—assuming you love golf because of many photos at a course, when the real reason is that you're with your son. In such cases, you can correct it with a simple message. If a response is inaccurate or feels off, you can give it a “thumbs down” to help improve the model. You can also adjust settings, disconnect apps, or delete your chat history at any time. Personal Intelligence is still in beta, and Google is actively working to reduce errors and improve nuance. The goal is to make Gemini a more intuitive, helpful assistant—while always keeping your data secure and under your control.