微软Bing推出生成式搜索：AI大模型重塑搜索新范式

Introduction: The Evolution of Search

Since the introduction of LLM-powered chat answers last year, Bing has been at the forefront of redefining how users interact with information online. Trusted by hundreds of millions, Bing has consistently served as a gateway to knowledge, answers, and exploration. Today, we are taking a significant leap forward by offering an early preview of our new generative search experience, currently being deployed to a small percentage of user queries. This initiative represents a core part of our ongoing mission to revolutionize the search paradigm.

自去年二月推出由大语言模型（LLM）驱动的必应聊天回答以来，我们一直在努力推动搜索领域的持续变革。必应持续获得数亿用户的信赖，帮助他们查找信息、解答问题并满足好奇心。今天，我们很高兴分享全新生成式搜索利用生成式人工智能技术优化搜索结果呈现方式，提供连贯、全面的答案而非简单链接列表的新型搜索模式。体验的早期预览，该功能目前正面向一小部分用户查询进行发布。这标志着我们在重塑搜索范式使命中迈出的关键一步。

What is Generative Search?

Bing's generative search integrates the capabilities of generative AI and large language models directly into the search results page. This fusion creates a bespoke, dynamic response tailored to each user's specific query. Rather than just presenting a list of links, Bing now synthesizes information to deliver a comprehensive, easy-to-understand answer directly on the page.

必应的生成式搜索利用生成式人工智能技术优化搜索结果呈现方式，提供连贯、全面的答案而非简单链接列表的新型搜索模式。将生成式人工智能和大语言模型的能力直接整合到搜索结果页面中。这种融合为每个用户的特定查询创建了定制化、动态的响应。必应不再仅仅呈现链接列表，而是综合信息，直接在页面上提供全面、易于理解的答案。

A Practical Example

For instance, when a user searches "What is a spaghetti western?", Bing generates an AI-powered experience that delves into the film subgenre. This response includes its history, origins, notable examples, and more. The information is presented in a clear, readable format, complete with links and citations that indicate the source of the information and allow users to explore topics in greater depth. Crucially, traditional web search results remain prominently displayed alongside this new AI-generated content.

例如，当用户搜索“什么是意大利式西部片？”时，必应会生成一个由AI驱动的体验，深入探讨这一电影子类型。该响应包括其历史、起源、著名示例等。信息以清晰、可读的格式呈现，并附有链接和引用，标明信息来源，方便用户进行更深入的探索。至关重要的是，传统的网页搜索结果仍然与此新的AI生成内容一同突出显示。

How It Works: Combining Scale and Precision

This new experience is built upon a dual foundation: the robust infrastructure of Bing's core web index and the advanced reasoning of language models, both large (LLMs) and small (SLMs). The system follows a sophisticated process:

Query Understanding: It interprets the intent and nuance behind the user's search query.
Source Synthesis: It reviews and draws connections across millions of information sources.
Dynamic Content Matching: It identifies and retrieves the most relevant facts and data.
AI-Generated Layout: It presents the synthesized information in a new, optimized format designed to fulfill the user's query more effectively than a list of links alone.

这一新体验建立在双重基础之上：必应核心网络索引的强大基础设施，以及大语言模型（LLM）和小语言模型（SLM）的高级推理能力。系统遵循一个复杂的流程：

查询理解：解读用户搜索查询背后的意图和细微差别。

来源综合：审查并关联数百万个信息源。

动态内容匹配AI系统实时分析查询，从数百万信息源中筛选并组合相关内容，生成个性化的搜索结果布局。：识别并检索最相关的事实和数据。

AI生成布局：以全新的优化格式呈现综合信息，旨在比单纯的链接列表更有效地满足用户的查询需求。

Commitment to Accuracy and a Healthy Web Ecosystem

Accuracy and trust remain our top priorities. We have applied rigorous refinements and optimizations developed for Bing's overall accuracy to this generative AI system. Furthermore, we are deeply mindful of the impact on the broader web ecosystem and content publishers.

准确性和可信度仍然是我们的首要任务。我们已将针对必应整体准确性开发的严格改进和优化措施应用于此生成式AI系统。此外，我们高度重视对更广泛的网络生态系统和内容发布者的影响。

Early data analysis is encouraging. The generative search experience is designed to maintain, and in some cases increase, the volume of clicks to publisher websites. Key design principles supporting this include:

Prominent Traditional Results: Standard search listings remain highly visible.
Increased Referential Links: The AI-generated answer incorporates multiple clickable references and citations, driving traffic to source material.
User Choice: Users can seamlessly choose between the concise AI summary and exploring the original sources.

We believe this approach supports a sustainable and vibrant internet where both users find answers efficiently and publishers receive valuable traffic.

早期数据分析令人鼓舞。生成式搜索利用生成式人工智能技术优化搜索结果呈现方式，提供连贯、全面的答案而非简单链接列表的新型搜索模式。体验的设计旨在维持甚至增加发布者网站的点击量。支持这一目标的关键设计原则包括：

突出的传统结果：标准搜索列表保持高度可见。

增加的参考链接：AI生成的答案包含多个可点击的参考和引用，将流量导向原始资料。

用户选择：用户可以无缝地在简洁的AI摘要和探索原始来源之间进行选择。
我们相信，这种方法有助于维持一个可持续且充满活力的互联网环境，既让用户高效找到答案，也让发布者获得有价值的流量。

A Measured Rollout and Invitation for Feedback

This represents a significant evolution for Bing, and we are committed to a responsible, iterative launch. We are rolling out this experience slowly, prioritizing learning and refinement over speed. User feedback is invaluable to this process.

这对必应来说是一次重要的演进，我们致力于负责任、迭代式的发布。我们正在缓慢推出此体验，优先考虑学习和改进，而非追求速度。用户反馈对此过程至关重要。

We invite you to share your thoughts directly within the search experience:

Use the thumbs up/thumbs down icons at the top of the generative answer.
Click the Feedback icon at the bottom of the search results page to provide detailed comments.

我们邀请您直接在搜索体验中分享您的想法：

使用生成答案顶部的点赞/点踩图标。

点击搜索结果页面底部的反馈图标以提供详细评论。

We will take our time to test, learn, and incorporate your insights as we work to create an exceptional search experience before considering a broader release. We look forward to sharing more updates on this journey in the coming months.

在考虑广泛发布之前，我们将花时间进行测试、学习并吸纳您的见解，努力打造卓越的搜索体验。我们期待在未来几个月内分享此旅程的更多更新。