内容来源：https://aibusiness.com/agentic-ai/gemini-scours-apps-personal-intelligence

内容总结：

谷歌近日为其生成式人工智能助手Gemini推出重大升级，新增“个人智能”功能。该功能通过深度整合用户授权的Gmail、相册、YouTube及搜索数据，可跨平台分析信息，提供高度个性化的答案。

例如，用户查询车辆轮胎更换时，系统能自动从相册提取车牌信息、从邮件获取车型规格，并结合旅行照片中的天气情况推荐适配轮胎。谷歌强调，所有数据关联均需用户手动授权并随时可关闭，且系统不会直接以用户数据训练模型，仅学习如何响应用户的指令定位相关信息。

目前该功能处于测试阶段，仅面向美国部分付费订阅用户开放，未来将逐步推广至更多地区和免费用户。谷歌同时表示，系统已对医疗等敏感话题设置防护机制，并允许用户选择非个性化回答。尽管存在误关联信息的可能性，但用户可通过反馈机制帮助优化系统。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI应用场景
要开启生成式AI之旅，首先应关注能够优化人类信息交互体验的领域。

这项新功能覆盖Gmail、相册、YouTube和搜索等产品，可即时回应用户提问。
谷歌正对Gemini进行重大升级，通过开放全生态数据权限实现更高程度的个性化服务。

个人智能系统能够调用Gmail、相册、YouTube和搜索中的信息，以更精准地处理用户查询。
虽然Gemini此前已能引用用户应用数据，但最新升级实现了跨越复杂数据源的逻辑推理能力——例如从邮件或照片中提取具体细节来回答问题。根据官方博客说明，它甚至能融合多源信息生成"量身定制"的答案。

谷歌实验室兼Gemini副总裁乔希·伍德沃德在博客中举例：他在商店排队时通过Gemini为家用车选购轮胎，系统通过相册图片提供车牌信息，检索Gmail获取车辆规格详情，甚至根据全家俄克拉荷马州旅行的相册照片推荐全天候轮胎选项。

该公司表示，个人智能系统将标注信息来源，同时用户可要求生成不含个性化元素的回答。针对医疗健康等敏感话题，系统已设置防护机制。

尽管有人可能对向谷歌开放此类数据心存顾虑，但伍德沃德强调该功能以隐私保护为前提设计："应用关联默认关闭，用户可自主选择开启、指定关联范围并随时关闭。"

他指出该功能并非直接训练用户数据，而是基于特定指令与Gemini的响应进行优化："简而言之，我们并非训练系统记忆您的车牌号，而是训练它在您需要时能准确定位。"

与其他智能体AI产品类似，该功能可能因无关主题的意外关联或对细微语义的理解偏差而产生错误。为此，谷歌鼓励用户通过"点踩"功能提交反馈。

个人智能系统目前在美国面向符合条件的谷歌AI Pro和AI Ultra订阅用户开放测试版，即将拓展至更多国家及免费用户群体。

该工具支持网页端、Android和iOS系统，兼容Gemini模型选择器中所有模型，目前仅限个人谷歌账户使用。
搜索中的AI模式功能也将于近期上线。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The new feature works across Gmail, Photos, YouTube and Search to answer a user's questions.
Google is launching a major upgrade to Gemini that allows greater personalization for users by giving it access to the entire ecosystem.
Personal Intelligence can utilize information contained in Gmail, Photos, YouTube and Search to better deal with a user's queries.
While Gemini already references information from a user's apps, the latest update marks a step forward in that it is able to reason across complex sources and take specific details from, for example, an email or photo to answer a question. Indeed, it can even combine these to provide "uniquely tailored" answers, according to a blog post.
Josh Woodward, vice president of Google Labs and Google Gemini, explained in a blog post how he used it to buy tires for his minivan as he waited in line at a store, with Gemini providing his vehicle's license plate number via a picture in Photos, details of its specification by searching Gmail and even suggesting all-weather tire options based on more images found in Photos of a family trip to Oklahoma.
Personal Intelligence will include sourcing, but it's also possible to request responses that have no element of personalization, with guardrails applied to sensitive topics such as healthcare, according to the company.
While some might balk at the idea of giving Google access to this level of data, Woodward said the feature was developed with privacy in mind. "Connecting your apps is off by default: you choose to turn it on, decide exactly which apps to connect to, and can turn it off anytime," he said.
He also pointed out that the feature is not trained directly on a user's data but on specific prompts and Gemini's response. "In short, we don't train our systems to learn your license plate number; we train them to understand that when you ask for one, we can locate it," he explained.
Like other agentic AI products, the feature has the potential to produce errors via connections inadvertently being made between unrelated topics or Gemini's difficulty in understanding nuance or context. To that end, user feedback in the form of a "thumbs down" is encouraged.
Personal Intelligence is available in beta to eligible Google AI Pro and AI Ultra subscribers in the U.S. Expansion to more countries and the free tier is coming soon.
The tool works across the web, Android and iOS operating systems and with all of the models in the Gemini model picker, but is only being offered for personal Google accounts.
Functionality in AI Mode in Search is also promised soon.