返回新聞
產品展示AI Understanding 簡報

Google to let users tune Discover with natural-language requests

Google says users will soon be able to tell Discover what topics and links they want to see more or less of, while new controls also personalize Search and Google News audio briefings.

5 min readRead the primary source
Source-provided image accompanying Google to let users tune Discover with natural-language requests
主要來源文件來源記錄
出版商
blog.google
來源連結
blog.googlehttps://blog.google/products-and-platforms/products/search/personalize-search-discover-news/
來源類型
主要文件-我們直接閱讀的官方公告、文件、文件或第一方頁面。
背景60 秒內了解這一點

從這裡開始

關鍵術語

記憶體(代理記憶體)
AI 代理程式跨步驟或會話使用儲存的上下文來提高連續性。
特點
模型用來進行預測的輸入變數。
測試一下自己人工智慧道德測驗

發生了什麼事

Google announced new personalization features for Search, Discover, and Google News. The company says users will soon be able to describe Discover preferences in their own words, select preferred publishers for greater visibility across Google surfaces, and customize audio news briefings in the Google News Android app.

Google also says people using the Google News app on Android can customize daily audio briefings by choosing specific topics. The briefings will include source attribution and links to full articles, according to the post. In this description, the user-facing choice is the ability to choose specific topics for daily audio briefings in the Google News Android app. The source ties that choice to the briefings themselves and does not describe a different control in this passage. It also states what listeners are to receive with those briefings: source attribution and links to full articles. Thus, the announcement presents topic selection, attribution, and links as parts of the same description. The claims here are Google's description of the feature, as reflected in the post.

Google adds that participating publishers in its news AI pilot program will contribute deeper dives into key stories. The announcement does not identify the participating publishers, describe the pilot's selection criteria, or state whether the audio controls are available to every Android user. The wording identifies deeper dives as a contribution from participating publishers, but it does not supply names or criteria for participation. Availability also remains unstated in the announcement. Those omissions are part of what the post leaves unresolved: it describes the contribution and the controls while withholding the specific information needed to determine the pilot's participants, its selection criteria, and the reach of the audio controls.

The post also contains AI-generated article summaries labeled experimental, which are separate from the company's description of the product features. That distinction matters within the announcement itself. The summaries are identified as AI-generated and experimental, while the other points describe the announced personalization features. The post therefore places the summaries alongside the product description without making them the same item as the audio controls, topic choices, publisher selections, or natural-language requests. The available description establishes that separation, but it does not add further detail about how the experimental summaries operate.

來源詳情: blog.google

為什麼這很重要

The changes give users more direct influence over recommendation and search experiences that can shape which publishers, topics, and AI-generated summaries they encounter. They may also affect how publishers reach readers, although Google provides no independent measurement of relevance, traffic, accuracy, or user outcomes.

Personalization can increase user control while also shaping the range of material a person encounters. A system that remembers requests may continue emphasizing a preference after the user's interest has changed, and a request for less of one subject could affect exposure to related reporting. These points describe the tension in the announcement: more direct control can coexist with a narrower range of material. The concern is not that a particular result has been demonstrated here, but that the remembered request and the request for less of one subject could influence what appears. The possible effect is on the range of material a person encounters, which is why user control and exposure need to be considered together.

Those are risks to examine, not established outcomes in this announcement. Google does not disclose in the source how competing preferences are resolved, whether important breaking news can override a user's settings, or what privacy controls govern stored requests and source selections. The announcement consequently leaves the relevant mechanisms unspecified. It does not say how one preference would be handled against another, whether important breaking news would take precedence over settings, or how stored requests and source selections would be governed. Each question follows directly from the controls and memory described in the draft, while the source provides no answer to those questions.

The changes may also affect how publishers reach readers, although Google provides no independent measurement of relevance, traffic, accuracy, or user outcomes. That limitation applies to the significance of the announcement as described here. The draft identifies possible effects on publishers and on the material users encounter, but it does not present measurements that establish the size or direction of any such effect. Relevance, traffic, accuracy, and user outcomes remain unmeasured in the source. The significance is therefore a question for examination rather than a measured result, and the available claims should remain tied to what Google announced.

Interactive Mechanism

互動機制:它實際上是如何運作的

以互動方式探索這項發展背後的基礎技術。

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
互動式概念檢查+10 Points
AI Ethics Quiz

Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?

接下來看什麼

The key questions are how broadly the features roll out, how Google interprets and remembers natural-language requests, how preferred-source choices affect rankings and AI summaries, and whether audio briefings preserve accurate attribution and links to complete reporting.

The audio pilot deserves scrutiny because Google says participating publishers will provide deeper dives into key stories while briefings include attribution and links. Reporting should examine whether the spoken summaries accurately represent the linked articles, identify the original publisher clearly, preserve updates and corrections, and provide enough context for listeners to understand uncertainty. These checks follow from the features Google describes. A deeper dive, spoken summary, attribution, and link each concern how the briefing relates to the complete reporting. The central issue is whether the material presented in audio remains connected to the linked articles and to the original publisher in a way that gives listeners the context identified above.

Google has not disclosed the pilot's participants, editorial workflow, review process, or performance results. Those unknowns will determine whether the new controls amount mainly to convenience or materially change how people discover and assess news. The missing information covers both who is involved and how the briefing material is handled. Participants, workflow, review process, and performance results would each bear on the questions raised by the pilot, but the announcement supplies none of them. Until those points are disclosed, the significance of the controls remains open between the convenience described in the draft and a material change in discovery and assessment.

The key questions are how broadly the features roll out, how Google interprets and remembers natural-language requests, how preferred-source choices affect rankings and AI summaries, and whether audio briefings preserve accurate attribution and links to complete reporting. These questions connect the product description to the unresolved details. Rollout concerns reach, interpretation and memory concern the requests, preferred-source choices concern rankings and AI summaries, and the audio questions concern attribution and links. Together, they define what should be watched as the features develop, while keeping the focus on the specific uncertainties already identified in the announcement.

相關指引和測驗

AI 倫理AI 的未來什麼是人工智慧?測試你所知道的—嘗試免費的人工智慧測驗在我們的詞彙表中尋找人工智慧術語
覺得有用嗎?