type
status
date
slug
summary
tags
category
icon
password
📢【新聞標題】
OpenAI Launches a General Purpose Agent in ChatGPT
OpenAI 在 ChatGPT 中推出通用 AI 代理
📰【摘要】
OpenAI is launching a new general purpose AI agent in ChatGPT, which the company says can complete a wide variety of computer-based tasks on behalf of users. OpenAI says the agent can automatically navigate a user’s calendar, generate editable presentations and slideshows, and run code.
OpenAI 正在 ChatGPT 中推出一種新的通用 AI 代理，該公司表示該代理可以代表使用者完成各種基於電腦的任務。 OpenAI 表示，該代理可以自動瀏覽使用者的日曆、生成可編輯的簡報和幻燈片，並執行程式碼。
🗝️【關鍵詞彙表】
📝 agent (n.)
- 代理、代理人
- 例句: OpenAI is launching a new general purpose AI agent in ChatGPT.
- 翻譯: OpenAI 正在 ChatGPT 中推出一種新的通用 AI 代理。
📝 navigate (v.)
- 導航、瀏覽
- 例句: OpenAI says the agent can automatically navigate a user’s calendar.
- 翻譯: OpenAI 表示，該代理可以自動瀏覽使用者的日曆。
📝 synthesize (v.)
- 合成、綜合
- 例句: Deep Research’s ability to synthesize information from dozens of websites.
- 翻譯: Deep Research 從數十個網站綜合資訊的能力。
📝 offload (v.)
- 卸載、轉移
- 例句: ChatGPT into an agentic product that can take actions and offload tasks for users.
- 翻譯: 將 ChatGPT 轉變為可以執行操作並為使用者轉移任務的代理產品。
📝 benchmarks (n.)
- 基準、評估標準
- 例句: The model underlying ChatGPT agent offers state-of-the-art performance on several benchmarks.
- 翻譯: ChatGPT 代理底層模型在多個基準測試上提供了最先進的效能。
📝 safeguards (n.)
- 安全措施、保障
- 例句: Activate new safeguards to mitigate these risks.
- 翻譯: 啟動新的安全措施以減輕這些風險。
📝 mitigate (v.)
- 減輕、緩和
- 例句: Activate new safeguards to mitigate these risks.
- 翻譯: 啟動新的安全措施以減輕這些風險。
✍️【文法與句型】
📝 be able to
- 說明: 表示能夠做某事，具有能力或許可。
- 翻譯: 能夠、可以
- 例句: OpenAI says users will be able to interact with the agent simply by prompting ChatGPT in natural language.
- 翻譯: OpenAI 表示，使用者只需以自然語言提示 ChatGPT 即可與代理互動。
📝 It remains to be seen
- 說明: 表示某事仍然有待觀察或驗證。
- 翻譯: 仍有待觀察、尚未可知
- 例句: While ChatGPT agent sounds impressive, it remains to be seen how capable it truly is in the real world.
- 翻譯: 雖然 ChatGPT 代理聽起來令人印象深刻，但它在現實世界中的能力如何仍有待觀察。
📖【全文與翻譯】
OpenAI is launching a new general purpose AI agent in ChatGPT, which the company says can complete a wide variety of computer-based tasks on behalf of users.
OpenAI 正在 ChatGPT 中推出一種新的通用 AI 代理，該公司表示該代理可以代表使用者完成各種基於電腦的任務。
OpenAI says the agent can automatically navigate a user’s calendar, generate editable presentations and slideshows, and run code.
OpenAI 表示，該代理可以自動瀏覽使用者的日曆、生成可編輯的簡報和幻燈片，並執行程式碼。
The tool, called ChatGPT agent, combines several capabilities from OpenAI’s previous agentic tools, including Operator’s ability to click around on websites [https://techcrunch.com/2025/01/23/openai-launches-operator-an-ai-agent-that-performs-tasks-autonomously/], as well as Deep Research’s ability to synthesize information [https://techcrunch.com/2025/02/02/openai-unveils-a-new-chatgpt-agent-for-deep-research/] from dozens of websites into a concise research report.
這個名為 ChatGPT 代理的工具，結合了 OpenAI 之前代理工具的幾種功能，包括 Operator 在網站上點擊的功能，以及 Deep Research 從數十個網站將資訊綜合為簡潔研究報告的能力。
OpenAI says users will be able to interact with the agent simply by prompting ChatGPT in natural language.
OpenAI 表示，使用者只需以自然語言提示 ChatGPT 即可與代理互動。
ChatGPT agent is rolling out on Thursday to subscribers to OpenAI’s Pro, Plus, and Team plans.
ChatGPT 代理將於週四向 OpenAI 的 Pro、Plus 和 Team 方案的訂閱者推出。
To activate the tool, users can select “agent mode” in ChatGPT’s dropdown menu of tools.
要啟用該工具，使用者可以在 ChatGPT 的工具下拉選單中選擇「代理模式」。
The launch of ChatGPT agent represents OpenAI’s boldest attempt yet to turn ChatGPT into an agentic product that can take actions and offload tasks for users, rather than just answering questions.
ChatGPT 代理的推出代表了 OpenAI 大膽的嘗試，將 ChatGPT 轉變為可以執行操作並為使用者轉移任務的代理產品，而不僅僅是回答問題。
In recent years, Silicon Valley companies including OpenAI, Google, and Perplexity have unveiled dozens of AI agents that have promised to do just that.
近年來，包括 OpenAI、Google 和 Perplexity 在內的矽谷公司已經推出了數十種 AI 代理，這些代理承諾可以做到這一點。
However, these early version AI agents have proven to struggle with complex tasks [https://techcrunch.com/2025/02/04/openais-operator-agent-helped-me-move-but-i-had-to-help-it-too/?_thumbnail_id=2593103], and they seem less compelling as products than the ultimate vision tech executives pitch around AI agents.
然而，這些早期版本的 AI 代理已被證明難以應對複雜的任務，而且與科技主管們圍繞 AI 代理所宣傳的最終願景相比，它們作為產品的吸引力似乎較小。
That said, OpenAI says ChatGPT agent is far more capable than its previous offerings.
儘管如此，OpenAI 表示 ChatGPT 代理比其先前的產品更強大。
The company’s new agent can access ChatGPT connectors [https://techcrunch.com/2025/06/04/chatgpt-introduces-meeting-recording-and-connectors-for-google-drive-box-and-more/], allowing users to connect apps like Gmail and GitHub so that the agent can find relevant information to your prompts.
該公司的新代理可以訪問 ChatGPT 連接器，允許使用者連接 Gmail 和 GitHub 等應用程式，以便代理可以找到與您的提示相關的資訊。
OpenAI says ChatGPT agent has access to a terminal, and it can use APIs to access certain apps.
OpenAI 表示，ChatGPT 代理可以訪問終端，並且可以使用 API 訪問某些應用程式。
OpenAI suggests that users can tap ChatGPT agent to “plan and buy ingredients to make Japanese breakfast for four,” as well as “analyze three competitors and create a slide deck.”
OpenAI 建議使用者可以利用 ChatGPT 代理來「計劃和購買為四人準備日式早餐的食材」，以及「分析三個競爭對手並創建幻燈片」。
These kinds of capabilities requires ChatGPT agent to parse through websites, plan a course of action, and use tools — much more complicated tasks than OpenAI has previously tried to tackle with agents.
這些功能需要 ChatGPT 代理解析網站、計劃行動方案並使用工具——這比 OpenAI 之前嘗試用代理處理的任務複雜得多。
The model underlying ChatGPT agent offers state-of-the-art performance on several benchmarks, according to OpenAI.
OpenAI 表示，ChatGPT 代理底層模型在多個基準測試上提供了最先進的效能。
The company says the ChatGPT agent model scores 41.6% on Humanity’s Last Exam (pass@1), a difficult test made up of thousands of questions across more than one hundred subjects.
該公司表示，ChatGPT 代理模型在「人類最後的考試」(Humanity’s Last Exam) (pass@1) 中得分為 41.6%，這是一個由數千個問題組成的難題，涵蓋一百多個科目。
That’s roughly double what OpenAI’s o3 and o4-mini scored on the test.
這大約是 OpenAI 的 o3 和 o4-mini 在測試中得分的兩倍。
On FrontierMath, one of the hardest known math benchmarks, OpenAI says ChatGPT agent scores 27.4% when it has access to tools, such as a terminal for code execution.
在 FrontierMath 這個已知的最難數學基準測試之一中，OpenAI 表示，當 ChatGPT 代理可以訪問工具時，例如用於程式碼執行的終端，它的得分為 27.4%。
The previous state-of-the-art score comes from o4-mini, which scored just 6.3%.
之前的最先進分數來自 o4-mini，它僅得分 6.3%。
OpenAI notes that it developed ChatGPT agent with safety in mind, largely because the product presents some newfound capabilities that could make it more dangerous in the hands of a bad actor.
OpenAI 指出，它在開發 ChatGPT 代理時考慮到了安全性，主要是因為該產品提供了一些新發現的功能，這些功能可能會使其在壞人手中更加危險。
OpenAI has previously warned that agentic models could present more dangerous capabilities.
OpenAI 之前曾警告說，代理模型可能會呈現更危險的功能。
In a safety report [https://cdn.openai.com/pdf/6bcccca6-3b64-43cb-a66e-4647073142d7/chatgpt_agent_system_card_launch.pdf] for ChatGPT agent, OpenAI says it’s designated the model as “high capability” in biological and chemical weapon domains, which is defined in OpenAI’s Preparedness Framework as a model with the ability to “amplify existing pathways to severe harm.”
在 ChatGPT 代理的安全報告中，OpenAI 表示，它已將該模型指定為在生物和化學武器領域具有「高能力」，這在 OpenAI 的「準備框架」中定義為一種具有「擴大現有途徑以造成嚴重傷害」能力的模型。
OpenAI notes that it does not have direct evidence of this, but it’s decided to take a precautionary approach and activate new safeguards to mitigate these risks.
OpenAI 指出，它沒有直接證據表明這一點，但它已決定採取預防措施並啟動新的安全措施以減輕這些風險。
The new safeguards for ChatGPT agent include a monitor that works in real time as users interact with the product.
ChatGPT 代理的新安全措施包括一個監視器，該監視器在使用者與產品互動時即時工作。
OpenAI says it runs a classifier across every prompt entered into ChatGPT agent, determining whether the request is related to biology.
OpenAI 表示，它對輸入 ChatGPT 代理的每個提示執行一個分類器，以確定請求是否與生物學相關。
If so, OpenAI runs ChatGPT agent’s response through a second monitor that determines whether the content could be used to evoke a biological threat.
如果是這樣，OpenAI 會通過第二個監視器運行 ChatGPT 代理的響應，以確定該內容是否可用於引起生物威脅。
OpenAI also says it disabled ChatGPT’s memory feature for this agent to prevent misuse.
OpenAI 還表示，它為此代理禁用了 ChatGPT 的記憶功能，以防止濫用。
In other parts of ChatGPT, OpenAI’s memory feature allows the chatbot to reference information from previous user chats.
在 ChatGPT 的其他部分，OpenAI 的記憶功能允許聊天機器人引用先前使用者聊天中的資訊。
However, OpenAI says bad actors could use the feature in ChatGPT agent to exfiltrate sensitive data through prompt injection attacks.
但是，OpenAI 表示，不良行為者可以使用 ChatGPT 代理中的此功能，通過提示注入攻擊來洩露敏感數據。
The company says it may revisit adding the feature in the future, however.
該公司表示，將來可能會重新考慮添加此功能。
While ChatGPT agent sounds impressive, it remains to be seen how capable it truly is in the real world.
雖然 ChatGPT 代理聽起來令人印象深刻，但它在現實世界中的能力如何仍有待觀察。
Until now, agent technology has proven relatively brittle when interacting with the real world.
到目前為止，代理技術在與現實世界互動時已被證明相對脆弱。
That said, OpenAI says it’s developed a more capable model that’s able to deliver on the promise of AI agents.
儘管如此，OpenAI 表示，它已經開發出一種更強大的模型，該模型能夠兌現 AI 代理的承諾。
This story was updated with more information.
此報導已更新更多資訊。
🔗【資料來源】
文章連結：https://techcrunch.com/2025/07/17/openai-launches-a-general-purpose-agent-in-chatgpt/