OpenAI 推出 Decisions API：用预设选项实现快速决策

简要说明Brief Description

介绍 Decisions APIIntroducing the Decisions API

目录Table of Contents

• 内容
  Content

内容Content

有时候，你的应用需要非常非常快速地做出明智决策，比如路由请求、对图像进行分类，或决定智能体下一步该做什么。今天，我们推出了 Decisions API。它能让模型在不到一秒的时间内作出响应，让交互几乎即时完成。Decisions API 基于 GPT-6 Luna 运行，使用起来非常简单：给它一段文本或一张图像，定义问题和可能的答案，然后在应用中使用结果。通过让模型只专注于少数几个选项，我们可以让它的速度提升近 10 倍，同时保留图像理解、广泛的语言支持和安全防护等能力。接下来，我们和 Charlie 一起看看它的实际应用，先从文本输入开始。谢谢你，Romain。我们已经发现，Decisions API 能为各种各样的应用带来帮助。Sometimes your app needs to make a smart decision really, really fast. Routing a request, classifying an image, or deciding what an agent should do next. Today, we're introducing the Decisions API. It lets the model respond in a fraction of a second, so interacting with it feels pretty much instant. The Decisions API runs on GPT-6 Luna. It's very simple. You give it a text or an image, you define the question and the possible answers, and you use the result in your app. By focusing the model on just a few choices, we can make it nearly 10 times faster while keeping capabilities like image understanding, broad language support, and safety protection. But let's see that in action with Charlie, starting with text input. Thanks, Romain. We're already finding a wide variety of applications that can benefit from the Decisions API.

假设你想处理一些非结构化文本。一个简单的做法是从一大段文字中自动提取表单字段，不过我们还可以进一步拓展。我们可以利用这些输入，将支持工单或新收到的销售请求转给合适的团队。在这里，我们可以看到多个示例：松散的用户输入被迅速归类为销售线索。太棒了。而且顺便说一下，这段演示并没有加速。服务器端的 Decisions API 响应时间不到 100 毫秒，非常令人印象深刻。现在，我觉得 Decisions API 更令人兴奋的一点是，你还可以输入图像，让它基于视觉理解快速采取行动。来看看这个简单的例子。我有一个小游戏，Decisions API 会获取道路和障碍物的画面帧，并迅速判断汽车应该留在当前车道，还是向左或向右行驶。Let's say you want to work with some unstructured text. A simple thing you could do would be to automatically extract form fields from a long paragraph, but we can expand on that. We can take those inputs and use them to route support tickets or inbound sales requests to the right teams. Here, we can see multiple examples of loose user input being quickly classified into sales leads. That's awesome. And this is not sped up, by the way. On the server side, the Decisions API here responded in less than 100 milliseconds. Very impressive. Now, what I find even more exciting about the Decisions API is how you can also use image inputs to take a fast action based on visual understanding. So check out this simple example here. I have a quick game and the Decisions API is taking frames of the road and obstacles, quickly choosing if the car should stay in its lane or go left or right.

现在，如果我们就在这里加快速度，Decisions API 仍然能够把车引导到正确的车道，耗时却只有推理模型的一小部分。太酷了。所以你完全可以想象，将这个 API 与一些基础的计算机操作任务结合起来，比如截取屏幕截图并快速发送操作指令。当然，如果你要处理非常复杂的计算机操作或浏览器操作场景，仍然会需要 Astra 的能力。不过 Charlie，你也一直在将 Decisions API 与语音这种模态结合使用。它们搭配起来效果如何？ 是的，我发现使用 Decisions API 可以让实时交互显得更加灵敏。这里我把 GPT-Live-1 与一个动画角色搭配使用。GPT-Live-1 负责语音，而 Decisions API 则根据我说话的内容，选择角色应该使用哪种表情。Now, if we turn up the speed right there, the Decisions API is still able to steer the car to the correct lane at the fraction of the time a reasoning model would take. That's so cool. So you can really imagine pairing this API with some basic computer use tasks, taking screenshots and quickly sending actions back. Of course, you'll still want Astra capabilities if you have a very sophisticated computer use or browser use scenario. But Charlie, you've also been using Decisions API together with voice as a modality. How does that pair together? Yeah, I found that using the Decisions API can make live interactions feel even more responsive. Here I've paired GPT-Live-1 with an animated character. GPT-Live-1 handles the voice while the Decisions API is choosing which expression the character should use as I talk.

我们来试试看。嘿，最近怎么样？挺好的。我还撑得住，尽量保持轻松舒适。你呢？这一周简直忙疯了。哦，天哪。听起来真让人筋疲力尽。想聊聊最近发生了什么吗？哦，不是那种糟糕的忙，是好事连连，忙得不可开交。我们今天要发布 Decisions API，很多人都特别期待能用上它。太令人兴奋了。我知道很多开发者一直盼着能用它来构建应用。没错。这个应用会从我们提供的一组表情中进行选择，那些小小的反应可以和助手实际说的话一起，成为对话的一部分。不过，Romain，你一直在打造一种互动体验，还涉及机器人技术，对吧？确实如此。正如你所见，Hugging Face 的朋友们好心借给我们一台即将推出的可编程机器人 Microduck 预览机。Let's try it out. Hey, how's it going? Pretty good. I'm hanging in there, keeping it calm and cozy. How are you doing? It has been a crazy week. Oh, wow. That sounds exhausting. Want to talk about what's been going on? Oh, no. It's been crazy in a good way. We're launching the Decisions API today and a ton of people are so excited to have it. That's so exciting. I know a lot of developers have been looking forward to building with it. Absolutely. The app is choosing from a set of expressions we've given it and those little reactions can become part of the conversation alongside what the assistant is actually saying. But, Romain, you've been building an interactive experience that touches on robotics, right? I have indeed. As you can see, our friends at Hugging Face kindly lended us a preview unit of the upcoming programmable robot Microduck.

这个的名字叫 Lavender。我还集成了 GPT-Live，为它赋予语音智能；也集成了 Decisions API，帮助机器人根据摄像头画面决定该看向哪里。在这个演示中，选择很简单：向机器人发送指令，让它朝任一方向转动头部，看向正确的物体。Lavender，跟着苹果走。你现在看到的是，摄像头每隔几帧拍摄的画面都会由 Decisions API 进行分析，随后它会决定机器人应该看向哪里。没错，是 Decisions API 在判断哪个是苹果。确实如此。其实，如果你想试试稍微模糊一些的指令，尽管说。好的。Lavender，你能跟着水果走吗？没错。太棒了。现在，你将跟随概念背后的东西。And this one's name is Lavender. So, I also integrated GPT-Live to give it voice intelligence as well as the Decisions API to help the robot choose where to look based on its camera. So, for this demo, the choices are simple. It's about moving the robot's head in either direction to look at the right object by sending the command to the robot. Lavender, follow the apple. Now, what you're noticing here is that every few frames from the camera is being analyzed by the Decisions API, which in turn can take the decision as to where the robot should look. And it's the Decisions API that's deciding what's the apple. Absolutely. In fact, if you want to try with a different request that's a bit more ambiguous, feel free. Sure. Lavender, can you follow the fruit? Exactly. Amazing. You will now follow what is behind the concept.

很酷吧？现在，如果我们把第二个物体也放进画面里，并提出一个更模棱两可的问题，会怎么样呢？Lavender，哪个更好玩？好啦，现在 Lavender 会跟着游戏手柄走。因为显然，游戏手柄玩起来更有意思。要是我把它们调换位置呢？做得好，Lavender。说真的，Lavender 显然对我的游戏手柄比对苹果更感兴趣。对吧？如果我稍微往左移动一下，它就会跟着。我们就继续追踪游戏手柄吧。太棒了。很酷吧？这些只是几个快速演示，展示了 Decisions API 如何处理文本和视觉输入，并通过键盘或像这样通过语音进行交互。我们相信 Decisions API 将开启一种与 AI 互动的全新方式，已经迫不及待想看看大家会用它创造出什么了。It's pretty cool, right? Now, what if we bring a second object into the picture with a question that's even more ambiguous. Lavender, what's more fun to play with? There you go. Now, Lavender will follow the gamepad. Because naturally, your gamepad is more fun to play with. What if I flip them? Good job, Lavender. And really, Lavender is definitely more interested about my gamepad than about my apple, for sure. Right? If I go to the left a little bit, it will follow. And we'll just keep on tracking the gamepad instead. Incredible. Pretty cool, right? These were just a few quick demos of the Decisions API across text and vision inputs, driven either by keyboard or voice interactions like this one. We think the Decisions API will unlock a brand new way to interact with AI, and we can't wait to see what you all will build with it.

祝你构建顺利。Lavender，你对 Decisions API 感到兴奋吗？[Lavender 鸣叫]Happy building. Lavender, are you excited about the Decisions API? [Lavender chirps]