GPT-6 与面向所有人的智能 UI

上个月，我们为付费客户推出了首批 GPT‑6 模型，今天，我们通过一款新的 GPT‑6 模型，将这一代智能带给更多人，服务每周使用 ChatGPT 的超过 12 亿人。

ChatGPT 中的 GPT‑6 引入了 Intelligent UI，这是一项新能力，让 ChatGPT 能够通过完全交互式的用户界面快速作答，使日常答案更加可视化，让学习复杂主题更容易，并且能够即时创建工具来完成任务。

推出 Intelligent UI

我们训练 GPT‑6 使用文本、视觉元素和交互元素来组织回答，并根据你的问题选择它们如何组合在一起。借助 Intelligent UI，回答可以包含图形、可点击按钮、表单、图表，以及你可以直接在对话中使用的交互式体验。

格式取决于你所问的内容。比较可能最适合并排展示。解释可能通过交互式图表更容易理解。而当简单的文本回答是最有用的回应时，ChatGPT 也可以给你这样的回答。

为日常问题提供更可视化的回答

ChatGPT 可以通过展示计划如何衔接，让计划更容易遵循。周日烤肉的时间安排可以与食谱并列显示，或者公路旅行的停靠点可以出现在地图上，并附上哪些地方值得绕路的说明。

GPT‑5.6 Instant

你说：给我一个周日烤羊腿的计划，我要请朋友来，人数还在确定中，说实话

ChatGPT 说：

当然可以。既然人数还在变化，我会把整顿饭设计成只有羊肉真正需要按人数调整。大多数配菜都可以轻松多应付两三个人。

周日烤肉菜单

主菜： 大蒜、迷迭香和第戎烤羊腿
配菜： 脆烤土豆 + 烤胡萝卜、一大份绿叶沙拉，以及黄油薄荷豌豆
酱汁： 正宗羊肉肉汁，再加一份快速薄荷酱
甜点： 苹果酥配香草冰淇淋——容易提前做好，而且非常宽容

饮品方面，中等至饱满酒体的红葡萄酒，如里奥哈、西拉或赤霞珠，都很合适。

需要多少羊肉？

带骨羊腿是我招待朋友时的首选：有节日气氛，相对宽容，而且容易切。

客人	带骨羊肉	土豆
4	3–4 磅	2 磅
6	4–5 磅	3 磅
8	5–6 磅	4 磅
10	6–7½ 磅	5 磅

如果超过大约 8 个人，我会认真考虑两条较小的羊腿，而不是一条巨大的羊腿。它们更容易均匀烹饪，而且你可以让其中一条比另一条稍微更粉嫩一些。

羊肉

前一晚，慷慨地撒盐调味，并用橄榄油、第戎芥末、大量大蒜、切碎的迷迭香、黑胡椒和柠檬皮屑揉搓。冷藏过夜。

周日，在烘烤前大约一小时取出。以425°F 烤约 20 分钟，然后降至350°F，直到中心达到大约125–130°F 为三分熟，或135°F 为五分熟。温度计比按每磅计算分钟数可靠得多。

静置30–40 分钟，松松地盖上。这段较长的静置时间其实很有用：它让你有一个完全空闲的烤箱来把土豆烤脆并完成蔬菜。

土豆很重要

使用育空黄金土豆或 russet 土豆。去皮并切成大块，焯水约 8–10 分钟，沥干，然后在锅中摇晃，让边缘变粗糙。

用热橄榄油或鸭油烘烤，加入盐、迷迭香和整瓣大蒜。你要的是深金黄色、非常酥脆，而不只是煮熟。

低压力的周日时间表

假设你在下午 6:00 吃饭，我会这样安排一天：

早上： 做苹果酥、薄荷酱和沙拉酱。准备胡萝卜。土豆去皮切块，并浸泡在冷水中。
3:00： 羊肉从冰箱取出。
3:45 左右： 羊肉放入烤箱。确切时间取决于大小，所以使用温度计。
4:30： 土豆焯水并弄粗糙。
约 5:00： 羊肉取出静置。将烤箱调高到约 425°F，开始烤土豆。
5:15： 胡萝卜放入烤箱。
5:30： 用羊肉滴落的汁做肉汁。煮豌豆。最后一刻拌沙拉。
5:40： 切羊肉。
6:00： 开吃。
晚餐后： 把已经做好的苹果酥放回温热的烤箱烤 15–20 分钟，并配冰淇淋食用。

最大的优势是朋友陆续到来时，几乎没有什么需要照看。

如果你目前大概在6 到 10 人之间，我会按 8 人采购，并让灵活食材更充足。烤羊肉剩菜非常棒，所以稍微多一点远比周日发现不够要好。

如果你愿意，我可以：

帮我把烤羊腿计划调整为 5 位客人
为烤羊腿菜单添加素食配菜选项
就如何让烤羊腿计划成为无麸质提供建议

GPT‑6 Instant

更轻松地学习复杂主题

可视化和交互式解释让你探索某件事如何运作，改变一个输入看看会发生什么，或者一步步解决一个问题。

只需提问即可创建交互式体验

你也可以让 ChatGPT 为手头任务创建一个工具，例如用来探索储蓄如何增长的计算器、和朋友聚餐时分账的工具，或者你可以在对话中玩的游戏。

Intelligent UI 如何运作

借助 Intelligent UI，我们让 GPT‑6 自由地为每个问题创建最有用的界面，同时在网页和移动端保持快速且熟悉的体验。

我们构建了一个原生、可流式传输的组件库，以及一个在模型生成界面时处理该界面的编译器。该组件库为每个回答提供熟悉的设计基础，同时给模型留出空间来决定各部分如何组合。编译器让界面能够在模型生成时逐步出现，而不必等待整个回答完成。

我们扩展了训练方法，帮助模型在内容、布局、视觉元素和交互方面做出深思熟虑的决策。这包括评估它创建的界面是否清晰、有用和完整。GPT‑6 学会了使用组件库并做出良好的设计决策，包括如何清晰地组织信息、何时使用交互性，以及何时简单的文本回答就足够。

在改进模型的设计判断力和扩展其可创建内容方面仍有工作要做，但 Intelligent UI 标志着人们与 ChatGPT 互动方式的重大转变。ChatGPT 不再把每个问题都塞进同一种对话格式，而是可以结合 UI、数据和操作，创造全新的方式，让人们探索、理解信息并与信息互动。

前沿智能——更快交付

借助 GPT‑6，我们还让 ChatGPT 能够在继续思考的同时开始作答。

GPT‑5.6 Instant

GPT‑6 Instant

随着推理模型的引入，用户可以在更困难的问题上获得帮助，但必须等待模型完成思考才能得到答案。现在，ChatGPT 可以将思考与回答交错进行。我们训练它考虑用户的等待时间，并利用其截至当时的知识和发现逐步组织答案。它学会了跨多个部分回复构建答案，每个部分都添加有用的信息，没有不必要的填充内容，同时保持完整答案像一次性写成的答案一样连贯且符合事实。

在一项针对高价值日常智能体任务的内部评估中，GPT‑6 Extra High 能够在与 GPT‑5.6 Medium 相同的时间内开始回答，同时取得比 GPT‑5.6 Extra High 更好的总体得分。

对于需要网络搜索的问题，GPT‑6 Instant 平均比 GPT‑5.6 Instant 提前 44% 开始回答。

GPT‑6 还提高了需要网络搜索的答案质量。它能更好地判断何时需要查询，并更可靠地找到支持其答案的信息。在一项针对难题的内部评估中，GPT‑6 比 GPT‑5.6 更经常正确回应用户问题的关键方面。

安全部署 GPT‑6

随着我们将更智能的 AI 带给更多人，我们致力于改进它识别现实世界风险的方式、诚实回应的方式，以及尊重人们所设限制的方式。GPT‑6 在这些领域都取得了进展，并建立在 Astra 的多项安全进步之上，以更好地抵御绕过其安全训练的尝试、遵循保障措施，并相对于 GPT‑5.6 Sol 更清楚地说明它能做什么和不能做什么。我们根据现实世界使用中的经验教训更新了其安全训练，以加强对涉及网络攻击、生物威胁和暴力的高风险滥用的防护。在对抗性测试中，GPT‑6 对绕过其安全训练的尝试表现出更强的抵抗力，尤其是跨多轮适应的攻击。GPT‑6 还经过训练，能在更高风险场景中安全回应，同时避免对无害请求进行不必要的拒绝。这包括利用对话历史和相关上下文来识别可能无法从单个提示中看出的风险。

我们在整个训练过程中进行了对齐评估，评估诚实性和对安全限制的回应。GPT‑6 更善于识别并传达自身能力的局限，包括在缺乏回答问题所需信息或工具时。有关结果和我们审查的更多细节，请参阅系统卡⁠（在新窗口中打开）。

可用性

配备 Intelligent UI 的 GPT‑6 今天开始在 Chat 标签中面向 ChatGPT Plus、Pro、Business 和 Enterprise 层级全球逐步推出。从明天开始，推出范围将扩展到 Free 和 Go 层级。Enterprise 可用性取决于工作场所管理员设置。

ChatGPT 中的 GPT‑6 由面向 Plus、Pro、Business 和 Enterprise 层级的 GPT‑6 Sol 以及面向 Free 和 Go 层级的 GPT‑6 Luna 提供支持。两者都针对日常对话进行了调优。

此更新适用于 ChatGPT 中的 Chat 体验。为 Work 和 Codex 提供支持的模型不会作为此次发布的一部分发生变化。

下一步

我们的使命是将通用人工智能的好处安全地带给全人类。GPT‑6 将我们在智能方面的最新进展带给每周使用 ChatGPT 的超过 12 亿人。

Intelligent UI 是迈向一种能够围绕你试图完成的事情塑造界面的 ChatGPT 的一步。

随着时间的推移，我们希望 ChatGPT 能为日常任务创造更多人们需要的界面。你应该能够描述你想要完成的事情，并围绕它构建一种体验来使用。

几十年来，人们不得不学习如何使用软件，在围绕特定任务设计的固定界面中导航。我们相信未来会有所不同。不是人们适应软件，而是软件适应人们。

Last month we introduced the first GPT‑6 models for paid customers and today we're bringing that next generation of intelligence to more people with a new GPT‑6 model built for more than 1.2 billion people who use ChatGPT each week.

GPT‑6 in ChatGPT introduces Intelligent UI, a new capability that allows ChatGPT to answer quickly with fully interactive user interfaces that make everyday answers more visual, make learning complex topics easier and that can quickly create a tool to solve a task right in the moment.

Introducing Intelligent UI

We’ve trained GPT‑6 to compose responses using text, visuals and interactive elements, choosing how they fit together based on your question. With Intelligent UI, responses can include graphics, tappable buttons, forms, charts, and interactive experiences you can use directly in your conversation.

The format depends on what you’re asking. A comparison might work best side by side. An explanation might be easier to understand through an interactive diagram. And when a simple text answer is the most useful response, ChatGPT can give you that.

Sunday roast menu

Main: Garlic, rosemary & Dijon roast leg of lamb
Sides: Crispy roast potatoes + roasted carrots, a big green salad, and peas with butter & mint
Sauce: Proper lamb gravy plus a quick mint sauce
Dessert: Apple crumble with vanilla ice cream — easy to make ahead and very forgiving

For drinks, a medium/full-bodied red such as Rioja, Syrah or Cabernet works nicely.

How much lamb?

A bone-in leg is my pick for having friends over: festive, relatively forgiving, and easy to carve.

Guests	Bone-in lamb	Potatoes
4	3–4 lb	2 lb
6	4–5 lb	3 lb
8	5–6 lb	4 lb
10	6–7½ lb	5 lb

If you get beyond about 8 people, I'd seriously considertwo smaller legs rather than one enormous one. They're easier to cook evenly, and you can leave one slightly pinker than the other.

The lamb

The night before, season it generously with salt and rub it witholive oil, Dijon, lots of garlic, chopped rosemary, black pepper and lemon zest. Refrigerate overnight.

On Sunday, take it out roughly an hour before roasting. Roast at425°F for about 20 minutes, then reduce to350°F until the center reaches roughly125–130°F for medium-rare or135°F for medium. A thermometer is much more reliable than calculating minutes per pound.

Rest it 30–40 minutes, loosely covered. That long rest is actually useful: it gives you a completely free oven for crisping the potatoes and finishing the vegetables.

The potatoes are important

Use Yukon Golds or russets. Peel and cut them into large chunks, parboil for about 8–10 minutes, drain, then shake them around in the pot to rough up the edges.

Roast them in hot olive oil or duck fat with salt, rosemary and whole garlic cloves. You want them deeply golden and aggressively crispy rather than merely cooked.

A low-stress Sunday schedule

Assuming you're eating at 6:00 pm, I'd run the day like this:

Morning: Make the crumble, mint sauce and salad dressing. Prep carrots. Peel/cut potatoes and leave them submerged in cold water.
3:00: Lamb out of refrigerator.
3:45-ish: Lamb into oven. Exact timing depends on its size, so use the thermometer.
4:30: Parboil potatoes and rough them up.
~5:00: Lamb comes out to rest. Turn oven up to about 425°F and get potatoes roasting.
5:15: Carrots into oven.
5:30: Make gravy from the lamb drippings. Cook peas. Toss salad at the very last minute.
5:40: Carve lamb.
6:00: Eat.
After dinner: Put the already-made crumble back into the warm oven for 15–20 minutes and serve with ice cream.

The big advantage is thatalmost nothing needs attention while your friends are arriving.

If you're currently somewhere between, say,6 and 10 people, I'd shop for 8 and keep the flexible ingredients generous. Roast-lamb leftovers are excellent, so being slightly over is much better than discovering on Sunday that you're short.

If you want, I can:

Help me adjust the lamb roast plan for 5 guests
Add vegetarian side dish options for the lamb roast menu
Advise on making the lamb roast plan gluten-free

GPT‑6 Instant

Learn complex topics more easily

Visual and interactive explanations let you explore how something works, change an input to see what happens, or work through a problem step by step.

Create interactive experiences just by asking

You can also ask ChatGPT to create a tool for the task at hand, such as a calculator to explore how savings could grow, a bill splitter for dinner with friends, or a game you can play in the conversation.

How Intelligent UI works

With Intelligent UI, we gave GPT‑6 the freedom to create the most useful interface for each question, while keeping the experience fast and familiar across web and mobile.

We built a library of native, streamable components, along with a compiler that processes the interface as the model generates it. The library gives each response a familiar design foundation while leaving the model room to decide how the pieces fit together. The compiler allows the interface to appear progressively as the model generates it, without waiting for the entire response to be complete.

We expanded our training methods to help the model make thoughtful decisions about content, layout, visuals, and interaction. This included evaluating the interfaces it creates for clarity, usefulness, and completeness. GPT‑6 learned to use the component library and make good design decisions, including how to organize information clearly, when to use interactivity, and when a simple text response is enough.

There's still work ahead to improve the model's design judgment and expand what it can create, but Intelligent UI marks a significant shift in how people interact with ChatGPT. Instead of fitting every question into the same conversational format, ChatGPT can combine UI, data, and actions to create entirely new ways for people to explore, understand, and interact with information.

Frontier Intelligence – delivered more quickly

With GPT‑6, we’re also making it possible for ChatGPT to begin answering while it continues to think.

GPT‑5.6 Instant

GPT‑6 Instant

With the introduction of reasoning models, users could get help with much harder problems, but had to wait for the model to finish thinking before getting an answer. Now, ChatGPT can interleave thinking with answering. We trained it to take into account the user’s waiting time and progressively compose an answer using its knowledge and findings up to that point. It learned to build answers across multiple partial responses, each adding useful information without unnecessary filler, while keeping the full answer as cohesive and factual as one written all at once.

In an internal evaluation of high-value, everyday agentic tasks, GPT‑6 Extra High is able to begin answering in the same amount of time as GPT‑5.6 Medium while achieving a better overall score than GPT‑5.6 Extra High.

For questions that need web search, GPT‑6 Instant starts answering 44% sooner, on average, than GPT‑5.6 Instant.

GPT‑6 also improves the quality of answers that require web search. It makes better decisions about when to look something up and more reliably finds information that supports its answer. In an internal evaluation of difficult problems, GPT‑6 correctly addressed the key aspect of the user’s question more often than GPT‑5.6.

Deploying GPT‑6 safely

As we bring more intelligent AI to more people, we work to improve how it recognizes real-world risks, how honestly it responds, and how well it respects the limits that people set. GPT‑6 makes progress in each of these areas, building on several of Astra’s safety advances to better resist attempts to bypass its safety training, follow safeguards, and be clearer about what it can and can’t do relative to GPT‑5.6 Sol. We updated its safety training based on lessons from real-world use to strengthen protections against high-risk misuse involving cyberattacks, biological threats, and violence. In adversarial testing, GPT‑6 showed stronger resistance to attempts to bypass its safety training, particularly attacks that adapt across multiple turns. GPT‑6 is also trained to respond safely in higher-risk scenarios while avoiding unnecessary refusals of harmless requests. This includes using conversation history and relevant context to recognize risks that may not be apparent from a single prompt.

We ran alignment evaluations throughout training, assessing honesty and responses to safety restrictions. GPT‑6 is better at recognizing and communicating the limits of its own capabilities, including when it lacks the information or tools needed to answer. More detail on the results and our review is available in the system card⁠(opens in a new window).

Availability

GPT‑6 with Intelligent UI starts rolling out globally to ChatGPT Plus, Pro, Business and Enterprise tiers today in the Chat tab. Starting tomorrow, the rollout expands to Free and Go tiers. Enterprise availability depends on workplace admin settings.

GPT‑6 in ChatGPT is powered by GPT‑6 Sol for Plus, Pro, Business, and Enterprise tiers, and GPT‑6 Luna for Free and Go tiers. Both are tuned for everyday conversation.

This update applies to the Chat experience in ChatGPT. The models powering Work and Codex are not changing as part of this release.

What comes next

Our mission is to deliver the benefits of artificial general intelligence safely to all of humanity. GPT‑6 brings our latest advances in intelligence to the more than 1.2 billion people who use ChatGPT every week.

Intelligent UI is a step toward a ChatGPT that can shape an interface around what you’re trying to do.

Over time, we want ChatGPT to create more of the interfaces people need for everyday tasks. You should be able to describe what you want to accomplish and work with an experience built around it.

For decades, people have had to learn how to use software, navigating fixed interfaces designed around particular tasks. We believe the future will be different. Instead of people adapting to software, software will adapt to people.

本文内容采集自官方网站，排版和翻译可能与原页面存在差异。

阅读官方全文