TED演讲:chatgpt惊人潜力的内幕
字幕摘录
| 时间 | 英文 | 中文 |
|---|---|---|
| 0:04 | We started OpenAI seven years ago | 我们7年前就开始了OpenAI |
| 0:06 | because we felt like something really interesting was happening in AI, | 因为我们觉得爱尔有件有趣的事情 |
| 0:10 | and we wanted to help steer it in a positive direction. | 我们想帮助引导它朝着一个积极的方向发展。 |
| 0:15 | It's honestly just really amazing | 老实说,真的很棒 |
| 0:17 | to see how far this whole field has come since then. | 看看这整个领域从那时起有多远 |
| 0:20 | And it's really gratifying to hear from people like Raymond, | 听到雷蒙德这样的人的声音 实在令人欣慰 |
| 0:24 | who are using the technology we are building and others, | 他们正在使用我们正在建设的技术和其他人, |
| 0:27 | for so many wonderful things. | 这么多美好的事物。 |
| 0:29 | We hear from people who are excited, | 我们听到那些兴奋的人说 |
| 0:31 | we hear from people who are nervous, | 我们听到那些紧张的人说 |
展开字幕全文(627 条)
| 序号 | 英文 | 中文 |
|---|---|---|
| 1 | We started OpenAI seven years ago | 我们7年前就开始了OpenAI |
| 2 | because we felt like something really interesting was happening in AI, | 因为我们觉得爱尔有件有趣的事情 |
| 3 | and we wanted to help steer it in a positive direction. | 我们想帮助引导它朝着一个积极的方向发展。 |
| 4 | It's honestly just really amazing | 老实说,真的很棒 |
| 5 | to see how far this whole field has come since then. | 看看这整个领域从那时起有多远 |
| 6 | And it's really gratifying to hear from people like Raymond, | 听到雷蒙德这样的人的声音 实在令人欣慰 |
| 7 | who are using the technology we are building and others, | 他们正在使用我们正在建设的技术和其他人, |
| 8 | for so many wonderful things. | 这么多美好的事物。 |
| 9 | We hear from people who are excited, | 我们听到那些兴奋的人说 |
| 10 | we hear from people who are nervous, | 我们听到那些紧张的人说 |
| 11 | we hear from people who are concerned, | 我们从关心的人那里听到, |
| 12 | we hear from people who feel both those emotions at once. | 我们从那些同时感受这些情绪的人那里听到 |
| 13 | And honestly, that's how we feel. | 老实说,这就是我们的感受。 |
| 14 | Above all, it feels like we're entering an historic period right now, | 最重要的是,感觉我们正进入一个历史时期 |
| 15 | where we as a world are going to define a technology | 我们的世界将定义一种技术 |
| 16 | that will be so important for our society going forward. | 这对于我们社会前进将非常重要。 |
| 17 | And I believe that we can manage this for good. | 我相信我们可以永远管理好这件事。 |
| 18 | So today I want to show you the current state of that technology | 所以今天我想向大家展示这个技术的现状 |
| 19 | and some of the underlying design principles that we hold dear. | 以及我们珍视的一些基本设计原则。 |
| 20 | So the first thing I'm going to show you | 所以首先我要告诉你 |
| 21 | is what it's like to build a tool for an AI | 做人工智能工具的感觉 |
| 22 | rather than building it for a human. | 而不是为人类建造它。 |
| 23 | So we have a new Dolly model, which generates images, | 因此,我们有一个新的多利模型, 生成图像, |
| 24 | and we are exposing it as an app for chat GPT to use on your behalf. | 我们把它曝光为 聊天GPT 代表你使用的应用程序。 |
| 25 | And you can do things like ask, suggest a nice post-Ted meal | 你可以做一些事情,比如问, 建议一个漂亮的后Ted饭 |
| 26 | and draw a picture of it. | 并画其像. |
| 27 | Now, you get all of the sort of ideation and creative back and forth | 现在,你得到了所有的 那种想法 和创造性前后 |
| 28 | and taking care of the details for you that you get out of chat GPT. | 并照顾你的细节 你从聊天GPT。 |
| 29 | And here we go, it's not just the idea for the meal, | 我们开始吧 不只是吃饭的主意 |
| 30 | but a very, very detailed spread. | 但非常,非常详细的传播。 |
| 31 | So let's see what we're going to get. | 故当观之. |
| 32 | But chat GPT doesn't just generate images in this case. | 但聊天GPT并不只是在此情况下生成图像. |
| 33 | It doesn't just generate text, it also generates an image. | 它不仅生成文字,还生成一个图像. |
| 34 | And that is something that really expands the power | 这能真正扩大力量 |
| 35 | of what it can do on your behalf | 它能为你做什么 |
| 36 | in terms of carrying out your intent. | 就执行你的意图而言 |
| 37 | And I'll point out, this is all a live demo. | 我要指出,这都是现场演示。 |
| 38 | This is all generated by the AI as we speak, | 这都是我们发言时的人工智能造成的 |
| 39 | so I actually don't even know what we're going to see. | 所以我甚至不知道我们会看到什么。 |
| 40 | This looks wonderful. | 这看起来太棒了。 |
| 41 | Now I'm getting hungry just looking at it. | 现在,我饿了 只是看着它。 |
| 42 | Now, we've extended chat GPT with other tools, too. | 现在,我们已经扩展聊天GPT 其他工具,太。 |
| 43 | For example, memory. | 例如记忆. |
| 44 | You can, say, save this for later. | 你可以,说,留这个以后。 |
| 45 | And the interesting thing about these tools is they're very inspectable. | 这些工具的有趣之处在于 它们非常可检查。 |
| 46 | So you get this little pop-up here that says, use the Dolly app. | 所以,你得到这个小弹出 在这里说,用多莉的应用程序。 |
| 47 | This is coming to all chat GPT users over upcoming months. | 未来数月所有聊天的GPT用户都会收到这个消息。 |
| 48 | And you can look under the hood and see that what it actually did | 你可以看看引擎盖下面 看看它到底做了什么 |
| 49 | was write a prompt, just like a human could. | 写一个快速的, 就像人类可以。 |
| 50 | And so you sort of have this ability to inspect | 所以你有这种能力检查 |
| 51 | how the machine is using these tools, | 机器是如何使用这些工具的, |
| 52 | which allows us to provide feedback to them. | 这使得我们能向他们提供反馈。 |
| 53 | Now it's saved for later, | 现在它保存以后, |
| 54 | and let me show you what it's like to use that information | 让我告诉你使用这个信息是什么感觉 |
| 55 | and to integrate with other applications, too. | 并与其他应用程序融合。 |
| 56 | You can say, | 你可以说, |
| 57 | I now make a shopping list for the tasty thing I was suggesting earlier. | 我现在做了一个购物清单 上面有我之前建议的东西 |
| 58 | I'm going to make it a little tricky for the AI. | 我要让人工智能变得有点棘手 |
| 59 | And tweet it out for all the TED viewers out there. | 给所有TED观众发微博 |
| 60 | So if you do make this wonderful, wonderful meal, | 所以如果你做了这顿美妙的晚餐 |
| 61 | I definitely want to know how it tastes. | 我绝对想知道它的味道。 |
| 62 | But you can see that chat GPT is selecting all these different tools | 但你可以看到聊天的GPT正在选择所有这些不同的工具 |
| 63 | without me having to tell it explicitly which ones to use in any situation. | 无需我明确告诉它 在任何情况下要使用哪种。 |
| 64 | And this, I think, shows a new way of thinking about the user interface. | 我认为,这显示了一种对用户界面的新思维方式。 |
| 65 | Like, we are so used to thinking of, | 像,我们是如此习惯 想, |
| 66 | well, we have these apps, we click between them, | 好了,我们有这些应用程序, 我们点击他们之间, |
| 67 | we copy-paste between them, | 我们抄袭他们之间, |
| 68 | and usually it's a great experience within an app, | 通常它是一个伟大的经验 在应用程序, |
| 69 | as long as you kind of know the menus and know all the options. | 只要你知道菜单和所有选择 |
| 70 | Yes, I would like you to. | 是的,我希望你这样做。 |
| 71 | Yes, please. | 是的,谢谢。 |
| 72 | Always good to be polite. | 总是好礼貌。 |
| 73 | And by having this unified language interface | 通过这个统一的语言界面 |
| 74 | on top of tools, | 在工具之上, |
| 75 | the AI is able to sort of take away all those details from you, | 人工智能能从你身上拿走所有细节 |
| 76 | so you don't have to be the one who spells out every single sort of little piece | 所以,你不必是一个 谁拼出每一个类型的小作品 |
| 77 | {\fn黑体\fs22\bord1\shad0\3aHBE\4aH00\fscx67\fscy66\2cHFFFFFF\3cH808080}什么应该发生的 | |
| 78 | And as I said, this is a live demo, | 正如我所说 这是现场演示 |
| 79 | so sometimes the unexpected will happen to us. | 因此,有时意外会发生在我们身上。 |
| 80 | But let's take a look at the Instacart shopping list | 但让我们看看Instacart购物名单 |
| 81 | while we're at it. | 当我们在它。 |
| 82 | And you can see we sent a list of ingredients to Instacart. | 你可以看到我们把一份原料清单 送到Instacart。 |
| 83 | Here's everything you need. | 这是你需要的一切。 |
| 84 | And the thing that's really interesting | 有趣的东西 |
| 85 | is that the traditional UI is still very valuable. | 传统的UI仍然非常宝贵。 |
| 86 | Right, if you look at this, | 好吧,如果你看看这个, |
| 87 | you still can click through it and sort of modify the actual quantities. | 您仍然可以点击并修改实际数量。 |
| 88 | And that's something that I think shows that they're not going away, | 我认为这说明他们不会离开 |
| 89 | 传统的UIs. | |
| 90 | It's just we have a new, augmented way to build them. | 只是我们有了新的,强化的方法来建造它们. |
| 91 | And now we have a tweet that's been drafted for our review, | 现在我们有一条微博 已经起草我们的审查, |
| 92 | which is also a very important thing. | 这也是一件非常重要的事。 |
| 93 | We can click Run, and there we are. | 我们可以点击运行,我们到了。 |
| 94 | We're the manager, we're able to inspect, | 我们是经理 我们可以检查 |
| 95 | we're able to change the work of the AI if we want to. | 我们能改变AI的工作,如果我们想。 |
| 96 | And so after this talk, you will be able to access this yourself. | 所以,经过这次谈话之后, 你就可以自己访问这个。 |
| 97 | And there we go. | 就这样 |
| 98 | Cool. | 凉爽。 |
| 99 | So ... | 所以说... |
| 100 | Thank you, everyone. | 谢谢大家 |
| 101 | So we'll cut back to the slides. | 故切为滑. |
| 102 | Now, the important thing about how we build this, | 现在,重要的是我们如何建造这个, |
| 103 | it's not just about building these tools. | 不仅仅是建造这些工具 |
| 104 | It's about teaching the AI how to use them. | 乃教阿难如何用. |
| 105 | Like, what do we even want it to do | 像,我们甚至想要它做什么 |
| 106 | when we ask these very high-level questions? | 当我们问这些非常高级的问题时? |
| 107 | And to do this, we use an old idea. | 为了做到这一点,我们用了一个老主意。 |
| 108 | If you go back to Alan Turing's 1950 paper on the Turing test, | 如果你回到艾伦·图灵1950年关于图灵测试的论文 |
| 109 | he says, look, you'll never program an answer to this. | 他说,你看,你永远不会计划一个答案。 |
| 110 | Instead, you can learn it. | 相反,你可以学习它。 |
| 111 | You could build a machine like a human child | 你可以像人一样造一台机器 |
| 112 | and then teach it through feedback, | 然后通过反馈传授它, |
| 113 | have a human teacher who provides rewards and punishments | 有个能提供奖惩的老师 |
| 114 | as it tries things out and does things that are either good or bad. | 因为它会尝试一些事情 做一些好或坏的事情 |
| 115 | And this is exactly how we train ChatGPT. | 这就是我们如何训练ChatGPT的。 |
| 116 | It's a two-step process. | 两步过程. |
| 117 | First, we produce what Turing would have called a child machine | 首先,我们生产图灵所谓的儿童机器 |
| 118 | through an unsupervised learning process. | 通过一个不受监督的学习过程。 |
| 119 | We just show it the whole world, the whole internet, | 我们只是展示它 整个世界,整个互联网, |
| 120 | and say, predict what comes next in text you've never seen before. | 并说,预测接下来的文字 你从来没有见过。 |
| 121 | And this process imbues it with all sorts of wonderful skills. | 这个过程使它充满了各种绝妙的技巧。 |
| 122 | For example, if you're shown a math problem, | 例如,如果你被显示一个数学问题, |
| 123 | the only way to actually complete that math problem, | 是真正解决数学问题的唯一办法 |
| 124 | say, what comes next, that green nine up there, | 接下来会发生什么 上面那个绿色的9号 |
| 125 | is to actually solve the math problem. | 也就是解决数学问题。 |
| 126 | But we actually have to do a second step, too, | 但实际上我们也要做第二步, |
| 127 | which is to teach the AI what to do with those skills. | 教人工智能如何运用这些技能。 |
| 128 | And for this, we provide feedback. | 为此,我们提供反馈。 |
| 129 | We have the AI try out multiple things, give us multiple suggestions, | 我们有AI尝试 多种事情, 给我们许多建议, |
| 130 | and then the human rates them, says, this one's better than that one. | 然后人类给它们评分 说这个比那个好 |
| 131 | And this reinforces not just the specific thing that the AI said, | 这不仅仅是强化了 大赦国际说, |
| 132 | but very importantly, the whole process that the AI used to produce that answer. | 但非常重要的是,人工智能用来制作答案的整个过程。 |
| 133 | And this allows it to generalize, | 这使得它能够概括, |
| 134 | it allows it to teach to sort of infer your intent | 它可以教它推断出你的意图 |
| 135 | and apply it in scenarios that it hasn't seen before, | 把它应用到以前从未见过的情景中 |
| 136 | that it hasn't received feedback. | 还没收到反馈 |
| 137 | Now, sometimes the things we have to teach the AI | 有时候我们得教人工智能 |
| 138 | are not what you'd expect. | 不是你想的那样 |
| 139 | For example, when we first showed GPT-4 to Khan Academy, | 例如,当我们第一次向汗学院展示GPT-4时, |
| 140 | they said, wow, this is so great, | 他们说,哇,这太棒了, |
| 141 | we're going to be able to teach students wonderful things. | 我们会教学生一些美妙的东西 |
| 142 | Only one problem, it doesn't double-check students' math. | 只有一个问题,它不重复检查学生的数学。 |
| 143 | If there's some bad math in there, | 如果有坏的数学在那里, |
| 144 | it will happily pretend that one plus one equals three and run with it. | 它会很高兴地假装一加一等于三 并运行它。 |
| 145 | So we had to collect some feedback data. | 所以我们不得不收集一些反馈数据。 |
| 146 | Sal Khan himself was very kind | 沙尔汗本人很善良 |
| 147 | and offered 20 hours of his own time to provide feedback to the machine | 给了他20个小时的时间给机器提供反馈 |
| 148 | alongside our team. | 与我们的团队。 |
| 149 | And over the course of a couple months, | 几个月来 |
| 150 | we were able to teach the AI that, hey, | 我们可以教人工智能,嘿, |
| 151 | you really should push back on humans in this specific kind of scenario. | 你真的应该 推回人类 在这种特定的情景。 |
| 152 | And we've actually made lots and lots of improvements | 我们做了很多改进 |
| 153 | to the models this way, | 对模型这样, |
| 154 | and when you push that thumbs down in ChatGPT, | 当你把拇指往下推 在ChatGPT, |
| 155 | that actually is kind of like sending up a bat signal to our team | 这其实有点像 向我们的团队发出蝙蝠信号 |
| 156 | to say, here's an area of weakness where you should gather feedback. | 可以说,这里是一个薄弱环节, 你应该收集反馈。 |
| 157 | And so when you do that, | 所以,当你这样做, |
| 158 | that's one way that we really listen to our users | 这也是我们真正倾听用户意见的一种方式 |
| 159 | and make sure we're building something that's more useful for everyone. | 并且确保我们建造的东西 对每个人都更有用。 |
| 160 | Now, providing high-quality feedback is a hard thing. | 现在,提供高质量的反馈是件难事。 |
| 161 | If you think about asking a kid to clean their room, | 如果你考虑让孩子清理房间 |
| 162 | if all you're doing is inspecting the floor, | 如果你只是检查地板 |
| 163 | you don't know if you're just teaching them to stuff all the toys in the closet. | 你不知道你是否只是教他们 把玩具都塞进衣柜里 |
| 164 | This is a nice dolly-generated image, by the way. | 顺便说一句,这是个漂亮的多莉生成的图像。 |
| 165 | And the same sort of reasoning applies to AI. | 同样的推理也适用于AI. |
| 166 | As we move to harder tasks, | 当我们转向更困难的任务时, |
| 167 | we will have to scale our ability to provide high-quality feedback. | 我们必须扩大提供高质量反馈的能力。 |
| 168 | But for this, the AI itself is happy to help. | 但为此,AI本身乐于提供帮助. |
| 169 | It's happy to help us provide even better feedback | 很高兴能帮我们提供更好的反馈 |
| 170 | and to scale our ability to supervise the machine as time goes on. | 随着时间的流逝,我们有能力监督机器 |
| 171 | And let me show you what I mean. | 让我告诉你我的意思。 |
| 172 | For example, you can ask for a GPD-4 question like this, | 例如,你可以问一个GPD-4这样的问题, |
| 173 | of how much time passed between these two foundational logs | 这两个基础日志之间的时间间隔 |
| 174 | on supervised learning and learning from human feedback, | 关于从人类反馈中接受监督的学习和学习, |
| 175 | and the model says, two months passed. | 模特儿说两个月过去了 |
| 176 | But is it true? | 但这是真的吗? |
| 177 | Like, these models are not 100 percent reliable, | 这些模型不百分之百可靠 |
| 178 | although they're getting better every time we provide some feedback. | 虽然每一次我们提供反馈 他们都越来越好 |
| 179 | But we can actually use the AI to fact-check, | 但我们可以用人工智能来核实事实 |
| 180 | and it can actually check its own work. | 它可以检查自己的工作。 |
| 181 | You can say, fact-check this for me. | 你可以说,事实检查这个给我。 |
| 182 | Now, in this case, I've actually given the AI a new tool. | 在这种情况下,我给了AI一个新的工具。 |
| 183 | This one is a browsing tool, | 这个是浏览工具 |
| 184 | where the model can issue search queries and click into web pages. | 其中,模型可以发布搜索查询并点击网页。 |
| 185 | And it actually writes out its whole chain of thought as it does it. | 实际上,它写出了整个思想链 就像它一样。 |
| 186 | It says, I'm going to search for this, and it actually does the search. | 上面写着,我要去寻找这个, 它实际上做了搜索。 |
| 187 | It finds the publication date and the search results. | 它找到出版日期和搜索结果。 |
| 188 | It then is issuing another search query. | 然后它会发布另一个搜索查询。 |
| 189 | It's going to click into the blog post. | 将点击博客文章。 |
| 190 | And all of this you could do, but it's a very tedious task. | 所有这一切你可以做, 但这是一个非常乏味的任务。 |
| 191 | It's not a thing that humans really want to do. | 这不是人类真正想要做的事情。 |
| 192 | It's much more fun to be in the driver's seat, | 坐在驾驶座上更好玩 |
| 193 | to be in this manager's position, | 成为这个经理的职位, |
| 194 | where you can, if you want, triple-check the work. | 在那里你可以,如果你想, 三倍检查工作。 |
| 195 | And out comes citations, | 并传出许多迹象, |
| 196 | so you can actually go and very easily verify | 所以你可以去核实一下 |
| 197 | any piece of this whole chain of reasoning. | 整个推理链的任何部分 |
| 198 | And it actually turns out, two months was wrong, | 事实上,两个月是错的 |
| 199 | two months in one week. | 一周内两个月 |
| 200 | That was correct. | 这是正确的。 |
| 201 | And we'll cut back to the slide. | 然后我们再切回幻灯片 |
| 202 | And so, the thing that's so interesting to me about this whole process | 所以,我对这个过程很感兴趣 |
| 203 | is that it's this many-step collaboration | 就是这种多步骤的合作 |
| 204 | between a human and an AI, | 在人类和人工智能之间 |
| 205 | because a human using this fact-checking tool | 因为一个人使用这个事实检查工具 |
| 206 | is doing it in order to produce data for another AI | 这样做是为了为另一个大赦国际提供数据。 |
| 207 | to become more useful to a human. | 对人更有用 |
| 208 | And I think this really shows the shape of something | 我觉得这真的 显示的东西的形状 |
| 209 | that we should expect to be much more common in the future, | 我们期望将来会更加常见, |
| 210 | where we have humans and machines kind of very carefully and delicately designed | 在那里,我们有人类和机器 一种非常仔细和微妙的设计 |
| 211 | in how they fit into a problem and how we want to solve that problem. | 如何融入一个问题 以及如何解决问题 |
| 212 | We make sure that the humans are providing the management, | 确保人类提供管理 |
| 213 | the oversight, the feedback, | 监督,反馈, |
| 214 | and that the machines are operating in a way that's inspectable and trustworthy. | 而机器的操作方式 是可检查的和可信的。 |
| 215 | And together, we're able to actually create even more trustworthy machines. | 我们一起可以创造出更可靠的机器。 |
| 216 | And I think that over time, if we get this process right, | 我认为,随着时间的推移, 如果我们把这个过程正确, |
| 217 | we will be able to solve impossible problems. | 我们能解决不可能的问题 |
| 218 | And to give you a sense of just how impossible I'm talking, | 为了让你明白我所说的不可能的事 |
| 219 | I think we're going to be able to rethink almost every aspect | 我想我们可以重新思考 几乎所有方面 |
| 220 | of how we interact with computers. | 我们如何与计算机互动。 |
| 221 | For example, think about spreadsheets. | 例如,考虑电子表格。 |
| 222 | They've been around in some form since, | 他们一直以某种形式出现,从那时起, |
| 223 | let's say, 40 years ago with VisiCalc. | 假设说 40年前和维西卡尔克在一起 |
| 224 | And I don't think they've really changed that much in that time. | 我想他们当时并没有改变这么多 |
| 225 | And here is a specific spreadsheet of all the AI papers on the archive | 这是档案里所有AI文件的具体电子表格 |
| 226 | for the past 30 years. | 过去30年 |
| 227 | There's about 167,000 of them, | 大约有167,000个 |
| 228 | and you can see the data right here. | 你可以在这里看到数据。 |
| 229 | But let me show you the chat GPT take on how to analyze a data set like this. | 但是,让我告诉你聊天GPT 如何分析这样的数据集。 |
| 230 | So we can give chat GPT access to yet another tool, | 这样我们可以让聊天GPT访问另一个工具, |
| 231 | 这个是Python翻译 | |
| 232 | so it's able to run code just like a data scientist would. | 所以它能像数据科学家一样运行代码 |
| 233 | And so you can just literally upload a file | 所以你可以直接上传一个文件 |
| 234 | and ask questions about it, | 你应当询问, |
| 235 | and very helpfully, it knows the name of the file, | 并且非常有帮助, 它知道文件的名称, |
| 236 | and it's like, oh, this is CSV, a comma-separated value file. | 这就像,哦,这是CSV, 一个逗号分隔的值文件。 |
| 237 | I'll parse it for you. | 我帮你分析一下 |
| 238 | The only information here is the name of the file, | 这里唯一的信息是文件的名字, |
| 239 | the column names like you saw, | 你所看到的列名, |
| 240 | and then the actual data. | 然后是实际数据。 |
| 241 | And from that, it's able to infer what these columns actually mean. | 由此可以推断出这些柱子的真正含义。 |
| 242 | That semantic information wasn't in there. | 语义信息不在那里。 |
| 243 | It has to put together its world knowledge | 它必须汇集世界知识 |
| 244 | of knowing that, oh, yeah, archive is the site that people submit papers, | 知道,哦,是的,档案 是人们提交文件的网站, |
| 245 | and therefore, that's what these things are, | 因此,这就是这些东西, |
| 246 | and that these are integer values, | 这些是整数, |
| 247 | and so therefore, it's a number of authors in the paper. | 因此,论文中有不少作者。 |
| 248 | All of that, that's work for a human to do, | 所有这一切, 这是人类的工作, |
| 249 | and the AI is happy to help with it. | AI很乐意帮忙。 |
| 250 | Now, I don't even know what I want to ask, | 现在我甚至不知道我想问什么 |
| 251 | so fortunately, you can ask the machine, | 幸运的是 你可以问机器 |
| 252 | can you make some exploratory graphs? | 你能做一些探索性的图表吗? |
| 253 | And once again, this is a super high-level instruction | 再说一遍,这是超高级的指令 |
| 254 | with lots of intent behind it, | 背后有许多意图 |
| 255 | but I don't even know what I want, | 但我甚至不知道我想怎样 |
| 256 | and the AI kind of has to infer what I might be interested in. | 人工智能必须推断出我感兴趣的是什么 |
| 257 | And so it comes up with some good ideas, I think. | 因此,我想,它产生了一些好主意。 |
| 258 | So a histogram of the number of authors per paper, | 所以每张论文作者人数的直观图 |
| 259 | time series of papers per year, word cloud of the paper titles. | 每年的时序论文,论文标题的词云. |
| 260 | All of that, I think, will be pretty interesting to see. | 我认为,所有这些将是非常有趣的。 |
| 261 | And the great thing is, it can actually do it. | 伟大的事情是,它实际上可以做到这一点。 |
| 262 | Here we go, a nice bell curve. | 在这里,我们走了,一个漂亮的钟曲线。 |
| 263 | You see that three is kind of the most common. | 你看,这三 是一种最常见的。 |
| 264 | It's going to then write a ... | 然后写... |
| 265 | It's going to make this nice plot of the papers per year. | 每年的报纸会是这么好的 |
| 266 | Something crazy is happening in 2023, though. | 2023年发生了一件疯狂的事 |
| 267 | Looks like we were on an exponential, and it dropped off a cliff. | 看起来像我们在指数, 它掉下悬崖。 |
| 268 | What could be going on there? | 怎么会这样? |
| 269 | And by the way, all this is Python code you can expect. | 顺便说一下,这一切都是 Python 代码,你可以期待。 |
| 270 | And then we'll see the word cloud, | 后见字云. |
| 271 | so you can see all these wonderful things that appear in these titles. | 这样你就能看到这些名词中出现的这些美妙的东西. |
| 272 | But I'm pretty unhappy about this 2023 thing. | 但我对2023年的事情很不高兴 |
| 273 | It makes this year look really bad. | 这让今年看起来很糟糕。 |
| 274 | Of course, the problem is that the year's not over. | 当然,问题在于今年还没有结束。 |
| 275 | So I'm going to push back on the machine. | 所以我要推回机器。 |
| 276 | So April 13th is the cutoff date, I believe. | 因此,4月13日是截止日期,我相信。 |
| 277 | Can you ... | 你能... |
| 278 | So we'll see, this is a kind of ambitious one. | 我们来看看,这是种雄心勃勃的。 |
| 279 | So, you know, again, I feel like there was more I wanted out of the machine here. | 所以,你知道,再次,我觉得有 更多的我想离开这里的机器。 |
| 280 | I really wanted it to, like, notice this thing. | 我真的希望它,喜欢,注意到这件事情。 |
| 281 | Maybe it's a little bit of an overreach for it to have sort of inferred magically | 可能有点太过份了 有点神奇 |
| 282 | that this is what I wanted. | 这就是我想要的。 |
| 283 | But I inject my intent. | 但是,我注入了我的意图。 |
| 284 | I provide this additional piece of, you know, sort of guidance. | 我提供额外的,你知道的, 那种指导。 |
| 285 | And under the hood, the AI is just writing code again, | 在引擎盖下,人工智能又在写代码 |
| 286 | so if you want to inspect what it's doing, it's very possible. | 所以,如果你想检查 它在做什么,这是非常可能的。 |
| 287 | And now, I'm going to push back on the machine. | 现在,我要推回机器。 |
| 288 | It does the correct projection. | 它做正确的投影。 |
| 289 | And if you notice, it even updates the title. | 如果你注意到,它甚至更新标题。 |
| 290 | I didn't ask for that, but it knows what I want. | 我没有要求,但它知道我想要什么。 |
| 291 | Now, we'll cut back to the slide again. | 现在,我们再切回幻灯片。 |
| 292 | This slide shows a parable of how I think we, | 这个幻灯片展示了一个例子,我怎么认为我们, |
| 293 | you know, a vision of how we may end up using this technology in the future. | 未来我们如何使用这种技术 |
| 294 | A person brought his very sick dog to the vet, | 有人把他病重的狗带到兽医那里 |
| 295 | and the veterinarian made a bad call to say, | 兽医打了一个不好的电话说, |
| 296 | let's just wait and see. | 让我们等着瞧。 |
| 297 | And the dog would not be here today had he listened. | 而狗今天不会在这里 如果他听。 |
| 298 | In the meanwhile, he provided the blood test, | 与此同时,他提供了血液测试, |
| 299 | like the full medical records, to GPT-4, | 和完整的医疗记录一样,到GPT-4, |
| 300 | which said, I am not a vet, you need to talk to a professional. | 也就是说 我不是兽医 你需要跟专业人士谈谈 |
| 301 | Here are some hypotheses. | 这里有一些假设。 |
| 302 | He brought that information to a second vet who used it to save the dog's life. | 他把这个信息带到第二个兽医,他用它来拯救狗的生命. |
| 303 | Now, these systems, they're not perfect. | 这些系统并不完美。 |
| 304 | You cannot overly rely on them. | 你不能过分依赖他们。 |
| 305 | But this story, I think, shows | 但这个故事,我认为,显示 |
| 306 | that the human with a medical professional | 拥有医学专业的人 |
| 307 | and with CHAT GPT as a brainstorming partner | 作为集思广益的合作伙伴 |
| 308 | was able to achieve an outcome that would not have happened otherwise. | 能够取得否则就不会取得的结果。 |
| 309 | I think this is something we should all reflect on and think about | 我觉得我们都应该思考一下 |
| 310 | as we consider how to integrate these systems into our world. | 我们考虑如何将这些系统融入我们的世界。 |
| 311 | And one thing I believe really deeply | 还有一件事我深信不疑 |
| 312 | is that getting AI right is going to require participation from everyone. | 获得AI的权利需要每个人的参与。 |
| 313 | And that's for deciding how we want it to slot in, | 这是用来决定我们如何让它进入 |
| 314 | that's for setting the rules of the road for what an AI will and won't do. | 这是为AI会做和不会做的事 制定道路规则 |
| 315 | And if there's one thing to take away from this talk, | 如果有一件事能从这个演讲中消失 |
| 316 | it's that this technology just looks different, | 这种技术看起来不同 |
| 317 | just different from anything people had anticipated. | 和人们预料的不一样 |
| 318 | And so we all have to become literate. | 因此,我们都必须识字。 |
| 319 | And that's honestly one of the reasons we released CHAT GPT. | 老实说,这就是我们释放CHAT GPT的原因之一。 |
| 320 | Together, I believe that we can achieve the OpenAI mission | 我相信我们可以一起完成OpenAI任务 |
| 321 | of ensuring that artificial general intelligence | 确保人工一般情报 |
| 322 | benefits all of humanity. | 使全人类受益。 |
| 323 | Thank you. | 谢谢 |
| 324 | Greg. | 格莱格。 |
| 325 | Wow. | 珃 |
| 326 | I mean, I suspect that within every mind out here, | 我的意思是,我怀疑 在每一个头脑在这里, |
| 327 | apart from there's a feeling of reeling, | 除了有回旋的感觉, |
| 328 | I suspect that a very large number of people viewing this, | 我怀疑很多人在看这个 |
| 329 | you look at that and you think, oh my goodness, | 你看看这个,你觉得, 哦,我的上帝, |
| 330 | pretty much every single thing about the way I work, I need to rethink. | 几乎每一件事 关于我的工作方式, 我需要重新考虑。 |
| 331 | There's just new possibilities there. | 这里只有新的可能性。 |
| 332 | I mean, right, who thinks that they're having to rethink the way that we do things? | 我的意思是,对的,谁认为 他们不得不重新思考 我们做事的方式? |
| 333 | Yeah, I mean, it's amazing, but it's also really scary. | 是啊,我的意思是,这是惊人的,但它也是非常可怕的。 |
| 334 | So let's talk, Greg, let's talk. | 所以,让我们谈谈,格雷格,让我们谈谈。 |
| 335 | Absolutely. | 当然 |
| 336 | I guess my first question actually is just how the hell have you done this? | 我的第一个问题就是 你是怎么做到的? |
| 337 | OpenAI有几百名员工 | |
| 338 | Google has thousands of employees working on artificial intelligence. | 谷歌有数千名员工从事人工智能工作. |
| 339 | Why is it you who's come up with this technology that's shocked the world? | 为什么你们想出这种技术来,让世界震惊? |
| 340 | The truth is, we're all building on shoulders of giants. | 事实是,我们都在建设 在巨人的肩膀上。 |
| 341 | There's no question if you look at the compute progress, | 毫无疑问,如果你看看计算的进展, |
| 342 | the algorithmic progress, the data progress, | 算法进步 数据进步 |
| 343 | all of those are really industry-wide. | 这些都是真正的工业。 |
| 344 | 但我认为在OpenAI, | |
| 345 | we made a lot of very deliberate choices from the early days. | 我们从早期就做了很多非常周密的选择。 |
| 346 | And the first one was just to confront reality as it lays, | 第一个只是面对现实 |
| 347 | and that we just sort of like thought really hard | 而我们只是有点喜欢 深思熟虑 |
| 348 | about what is it going to take to make progress here? | 需要什么才能在这方面取得进展? |
| 349 | We tried a lot of things that didn't work, so you only see the things that did. | 我们尝试了很多不起作用的事情,所以你只看到了那些事情。 |
| 350 | And I think that the most important thing | 我认为最重要的是 |
| 351 | {\fn黑体\fs22\bord1\shad0\3aHBE\4aH00\fscx67\fscy66\2cHFFFFFF\3cH808080}一直以来 {\fn黑体\fs22\bord1\shad0\3aHBE\4aH00\fscx67\fscy66\2cHFFFFFF\3cH808080}我一直想找一群 彼此迥然不同的人 | |
| 352 | to work together harmoniously. | 和谐地合作 |
| 353 | Can we have the water, by the way, just brought here? | 顺便说一句 能把水拿过来吗? |
| 354 | I think we're going to need it. | 我想我们需要它。 |
| 355 | It's a dry amount of topic. | 题数干. |
| 356 | But isn't there something also, | 但是,是不是还有一样东西, |
| 357 | just about the fact that you saw something in these language models | 你从这些语言模型中看到一些东西 |
| 358 | that meant that if you continue to invest in them and grow them, | 这意味着如果你继续投资 种植它们, |
| 359 | that something at some point might emerge? | 某时某事可能会出现? |
| 360 | Yes. | 对 |
| 361 | And I think that ... | 我觉得... |
| 362 | I mean, honestly, I think the story there is pretty illustrative. | 我的意思是,说实话,我认为 故事是相当说明性的。 |
| 363 | I think that at a high level, deep learning, | 我认为,在高水平,深刻的学习, |
| 364 | we always knew that was what we wanted to be, was a deep learning lab. | 我们一直知道这就是我们想要的 是一个深层的学习实验室 |
| 365 | And exactly how to do it, I think that in the early days, we didn't know. | 确切地说,如何做到这一点, 我认为在早期,我们不知道。 |
| 366 | We tried a lot of things, | 我们尝试了很多事情, |
| 367 | and one person was working on training a model | 有一人正在训练一个模型 |
| 368 | to predict the next character in Amazon reviews, | 为了预测亚马逊评论中的下一个角色, |
| 369 | and he got a result, | 他得到了一个结果, |
| 370 | where this is a syntactic process. | 在其中这是一个协同过程。 |
| 371 | You expect the model will predict where the commas go, | 你预计模型会预测逗号的去向, |
| 372 | where the nouns and verbs are, | 名词和动词在哪里, |
| 373 | but he actually got a state-of-the-art sentiment analysis classifier out of it. | 但实际上他得到了一个最新的情绪分析 分类器出来。 |
| 374 | This model could tell you if a review was positive or negative. | 这个模型可以告诉你审查是正面还是负面. |
| 375 | I mean, today, we are just like, | 我的意思是,今天,我们只是喜欢, |
| 376 | oh, come on, anyone could do that. | 哦,来吧,任何人都可以做到这一点。 |
| 377 | But this was the first time that you saw this emergence, | 但这是你第一次看到这种出现 |
| 378 | this sort of semantics that emerged from this underlying syntactic process. | 这种语义 产生于这个基本的 合成过程。 |
| 379 | And there, we knew, you've got to scale this thing, | 在那里,我们知道, 你必须缩小这个东西, |
| 380 | you've got to see where it goes. | 你得看看它会往哪里走 |
| 381 | CAOKEY So I think this helps explain the riddle that baffles everyone | 考凯 所以我觉得这能解释 谜语让每个人困惑 |
| 382 | looking at this, | 看这个 |
| 383 | because these things are described as prediction machines, | 因为这些东西被描述为预测机器, |
| 384 | and yet what we're seeing out of them, | 然而我们所看到的, |
| 385 | it just feels impossible that that could come from a prediction machine, | 感觉这不可能来自预测机, |
| 386 | just the stuff you showed us just now. | 你刚才给我们看的东西 |
| 387 | And the key idea of emergence is that when you get more of a thing, | 出现的关键思想是 当你得到更多的东西, |
| 388 | suddenly different things emerge. | 突然间发生了不同的事情。 |
| 389 | It happens all the time, ant colonies, single ants run around. | 它经常发生,蚂蚁殖民地, 单蚁到处跑. |
| 390 | When you bring enough of them together, you get these ant colonies | 当你把足够多的蚂蚁聚在一起, 你就会得到这些蚂蚁殖民地 |
| 391 | that show completely emergent and different behavior. | 展现出全新的行为 |
| 392 | Or a city, where a few houses together is just houses together, | 或一个城市, 几个房子在一起 只是房子在一起, |
| 393 | but as you grow the number of houses, | 但随着你增加房屋数量, |
| 394 | things emerge, like suburbs and cultural centers and traffic jams. | 就像郊区和文化中心 交通堵塞 |
| 395 | Give me one moment for you, | 给我一分钟给你 |
| 396 | when you saw just something pop that just blew your mind, | 当你看到刚刚爆炸的东西 让你心碎, |
| 397 | that you just did not see coming. | 你只是没想到会来 |
| 398 | CAOKEY Yeah, well, so you can try this in chat.jpt. | 考凯 是啊,所以你可以试试这个在聊天。 |
| 399 | If you add 40-digit numbers, | 如果你加上40位数字 |
| 400 | CAOKEY 40-digit? | 考基40位数? |
| 401 | CAOKEY 40-digit numbers, the model will do it, | CAOKEY的40位数字,模型会做到这一点, |
| 402 | which means it's really learned a internal circuit for how to do it. | 也就是说它真的学会了 内部的电路如何做到这一点。 |
| 403 | And the really interesting thing is, | 而真正有趣的是, |
| 404 | actually, if you have an ad, like a 40-digit number | 实际上,如果你有一个广告, 像40位数字 |
| 405 | plus a 35-digit number, it'll often get it wrong. | 加上一个35位数字,它往往会弄错了. |
| 406 | And so you can see that it's really learning the process, | 所以,你可以看到, 它真的学习的过程, |
| 407 | but it hasn't fully generalized, right? | 但它还没有完全普遍化,对不对? |
| 408 | It's like, you can't memorize the 40-digit addition table. | 这就像,你不能记住 40 位加表。 |
| 409 | That's more add-ins than there are in the universe. | 这比宇宙中的加法还要多. |
| 410 | So it had to have learned something general, | 所以它不得不学到一些东西 将军, |
| 411 | but it hasn't really fully yet learned that, | 但它还没有完全知道, |
| 412 | oh, I can generalize this to adding arbitrary numbers of arbitrary lengths. | 哦,我可以概括一下,增加任意的长度。 |
| 413 | CAOKEY So what's happened here is that you've allowed it to scale up | 考凯 所以这里发生的事情是 你允许它扩大 |
| 414 | and look at an incredible number of pieces of text, | 且看看许多不可思议的文字, |
| 415 | and it is learning things that you didn't know | 学习你不知道的东西 |
| 416 | that it was going to be capable of learning. | 它将有能力学习。 |
| 417 | CAOKEY Well, yeah, and it's more nuanced, too, | 考凯 嗯,是的,它更细微,太, |
| 418 | because one science that we're starting to really get good at | 因为有一个科学,我们开始 真正获得好 |
| 419 | is predicting some of these emerging capabilities. | 正在预测这些新兴能力。 |
| 420 | And to do that, actually, one of the things | 为了做到这一点,实际上,一件事 |
| 421 | that I think is very undersung in this field is sort of engineering quality. | 我认为这个领域非常低沉 是一种工程质量。 |
| 422 | We had to rebuild our entire stack and get ... | 我们不得不重建我们的整个堆,并得到... |
| 423 | When you think about building a rocket, | 当你想到建造火箭时, |
| 424 | you know every tolerance has to be incredibly tiny. | 你知道每一个容忍 必须是惊人的微小。 |
| 425 | Same is true in machine learning. | 机器学习也是如此 |
| 426 | You have to get every single piece of the stack engineered properly, | 你得把堆栈的每一块都妥善地设计好 |
| 427 | and then you can start doing these predictions. | 然后你可以开始做这些预测。 |
| 428 | There are all these incredibly smooth scaling curves | 这些都是非常平滑的曲线 |
| 429 | that tell you something deeply fundamental about intelligence. | 告诉你一些关于智慧的根本问题 |
| 430 | If you look at our GPT-4 blog post, you can see all of these curves in there. | 看看我们的GPT-4博客文章, |
| 431 | And now we're starting to be able to predict. | 现在我们开始可以预测了。 |
| 432 | So we were able to predict, for example, the performance on coding problems | 所以我们可以预测,比如,编码问题的表现 |
| 433 | from, you know, we basically look at some models | 从,你知道, 我们基本上看一些模型 |
| 434 | that are 10,000 times or 1,000 times smaller. | 即1万倍或1000倍较小。 |
| 435 | And so there's something about this that is actually smooth scaling, | 所以有些东西是平滑的 |
| 436 | even though it's still early days. | 虽然现在还很早 |
| 437 | CAOKEY So here is one of the big fears, then, that arises from this. | 考凯 因此,这里有一个很大的恐惧, 由此而来。 |
| 438 | If it's fundamental to what's happening here, | 如果这是这里发生的一切的根本 |
| 439 | that as you scale up, things emerge | 当你放大时,事物就会浮现 |
| 440 | that you can maybe predict in some level of confidence, | 你可以在某种程度的自信下预测 |
| 441 | but it's still capable of surprising you. | 但它仍然可以让你惊讶。 |
| 442 | Why isn't there just a huge risk of something truly terrible emerging? | 为何不冒着巨大的风险 出现真正可怕的事情? |
| 443 | CAOKEY Well, I think all these are questions of degree and scale and timing. | 考凯 我想这些都是程度、规模和时机的问题。 |
| 444 | And I think one thing people miss, too, is sort of the integration with the world. | 我认为,人们也忽略了一件事, 就是与世界融合。 |
| 445 | There's also this incredibly emergent, very powerful thing, too. | 还有一件令人难以置信的,非常强大的事情。 |
| 446 | And so that's one of the reasons | 所以这就是原因之一 |
| 447 | that we think it's so important to deploy incrementally. | 我们认为渐进部署非常重要 |
| 448 | And so I think that what we kind of see right now, | 所以我认为我们现在看到的, |
| 449 | if you look at this talk, a lot of what I focus on | 如果你看这段话,我关注的很多 |
| 450 | is providing really high-quality feedback. | 正在提供真正高质量的反馈。 |
| 451 | Today, the tasks that we do, you can inspect them, right? | 今天,我们的任务, 你可以检查,对不对? |
| 452 | It's very easy to look at that math problem | 看那个数学问题很容易 |
| 453 | and be like, no, no, no, machine seven was the correct answer. | 7号机是正确的答案 |
| 454 | But even summarizing a book, that's a hard thing to supervise. | 但是,即使总结一本书,这是很难监督的。 |
| 455 | How do you know if this book summary is any good? | 你怎么知道这本书的总结有什么用? |
| 456 | You have to read the whole book. | 须读全书. |
| 457 | No one wants to do that. | 没有人想这样做。 |
| 458 | And so I think that the important thing will be | 所以我认为重要的是 |
| 459 | that we take this step-by-step | 我们一步一步走 |
| 460 | and that we say, OK, as we move on to book summaries, | 而我们说,OK, 当我们开始书写摘要, |
| 461 | we have to supervise this task properly. | 我们必须妥善监督这项任务。 |
| 462 | We have to build up a track record with these machines | 我们必须用这些机器建立记录 |
| 463 | that they're able to actually carry out our intent. | 他们能够真正实现我们的意图 |
| 464 | And I think we're going to have to produce even better, more efficient, | 我认为我们必须生产出更好的,更有效的, |
| 465 | more reliable ways of scaling this, | 以更可靠的方式扩大范围, |
| 466 | sort of like making the machine be aligned with you. | 有点像制造机器 与你一致。 |
| 467 | CAOKEY NAPIER-MURPHY-MARTINEZ So we're going to hear later in the session, | 考基·纳尼尔-穆里普西-马蒂内斯 所以,我们将听到 在会后, |
| 468 | there are critics who say that there's no real understanding inside the system. | 有批评者说系统内部没有真正的理解. |
| 469 | Is it going to always ... | 它会永远... |
| 470 | We're never going to know that it's not generating errors, | 我们永远不会知道 它不会产生错误, |
| 471 | that it doesn't have common sense and so forth. | 它没有常识等等。 |
| 472 | Is it your belief, Greg, that that is true at any one moment, | 格雷格,你相信 任何时刻都是真的吗? |
| 473 | but that the expansion of the scale and the human feedback | 但规模的扩大和人类的反馈 |
| 474 | that you talked about | 你说过的 |
| 475 | {\fn黑体\fs20\shad2\2aH82\3aH20\4aH33\fscx95\3cH592001\be1}基本上要踏上那段旅程 | |
| 476 | of actually getting to things like truth and wisdom and so forth | 获得真理 智慧等等 |
| 477 | with a high degree of confidence? | 信心十足吗? |
| 478 | Can you be sure of that? | 你能确定吗? |
| 479 | CAOKEY NAPIER-MURPHY-MARTINEZ Yeah, well, I think that the opening eye ... | 考基·纳尼尔-穆里普西-马蒂内斯 是啊,我想开眼... |
| 480 | Yes, I believe that is where we're headed. | 是的,我相信这就是我们要去的地方。 |
| 481 | And I think that the opening-eye approach here | 我认为这里的开眼方法 |
| 482 | has always been just like, let reality hit you in the face. | 一直只是一样, 让现实打击你的脸。 |
| 483 | It's like this field is the field of broken promises | 这就像这个领域 是违背诺言的领域 |
| 484 | of all these experts saying X is going to happen, Y is how it works. | 这些专家说,X将会发生,Y就是这样的。 |
| 485 | People have been saying neural nets aren't going to work for 70 years. | 人们一直说神经网70年都没用 |
| 486 | They haven't been right yet. | 他们还没说对呢 |
| 487 | They might be right, maybe 70 years plus one or something like that | 他们也许是对的,也许70年 加上一个或者类似的东西 |
| 488 | is what you need. | 这是你需要的。 |
| 489 | But I think that our approach has always been, | 但我认为我们的方法一直是, |
| 490 | you've got to push to the limits of this technology | 你得把技术推向极限 |
| 491 | to really see it in action, | 真正看到它的行动, |
| 492 | because that tells you then, | 因为那告诉你, |
| 493 | oh, here's how we can move on to a new paradigm. | 哦,这是我们如何 前进到一个新的范式。 |
| 494 | And we just haven't exhausted the fruit here. | 我们只是还没把水果用完 |
| 495 | CAOKEY NAPIER-MURPHY-MARTINEZ It's quite a controversial stance you've taken, | 考基·纳尼尔-穆里普西-马蒂内斯 你的立场相当有争议 |
| 496 | that the right way to do this is to put it out there in public | 这样做的正确方式是公开发表 |
| 497 | and then harness all this. | 然后把这一切。 |
| 498 | Instead of just your team giving feedback, | 不只是你的团队提供反馈, |
| 499 | the world is now giving feedback. | 世界现在正在提供反馈。 |
| 500 | But if bad things are going to emerge, | 但是,如果坏事要发生, |
| 501 | it is out there. | 它就在这里。 |
| 502 | So the original story that I heard on OpenAI | 我在OpenAI上听到的原创故事 |
| 503 | when you were founded as a nonprofit, | 当时,你们原是非营利的。 |
| 504 | where you were there as the great sort of check | 在那里你在那里 作为伟大的检查类型 |
| 505 | on the big companies doing their unknown, possibly evil thing with AI, | 在大公司 做他们未知的, 可能是邪恶的事情与AI, |
| 506 | and you were going to build models that sort of somehow held them accountable | 而你打算建立一些模式 以某种方式追究他们的责任 |
| 507 | and was capable of slowing the field down, if need be. | 并且能够 放慢球场 如有必要。 |
| 508 | Or at least that's kind of what I had. | 或者至少这就是我拥有的 |
| 509 | And yet what's happened, arguably, is the opposite, | 然而,发生的事情,可以说是相反的, |
| 510 | that your release of GPT, especially chat GPT, | 你释放GPT, 特别是聊天GPT, |
| 511 | puts such shockwaves through the tech world | 科技界的冲击波 |
| 512 | 现在Google和Meta等等都在赶时间 | |
| 513 | And some of their criticisms have been, | 他们的一些批评是, |
| 514 | you are forcing us to put this out here without proper guardrails, | 你强迫我们把这个放在这里 没有合适的护栏 |
| 515 | or we die. | 否则我们就死定了 |
| 516 | How do you make the case | 你怎么解释? |
| 517 | that what you have done is responsible here and not reckless? | 你所做的一切 责任在这里,而不是鲁莽? |
| 518 | CAOKEY NAPIER-MURPHY-MARTINEZ Yeah, we think about these questions all the time, | 考基·纳尼尔-穆里普西-马蒂内斯 我们一直在想这些问题 |
| 519 | like seriously all the time, | 像一直以来认真的, |
| 520 | and I don't think we're always going to get it right. | 我不认为我们总是会得到它的权利。 |
| 521 | But one thing I think has been incredibly important, | 但我认为有件事很重要 |
| 522 | from the very beginning, | 從一開始, |
| 523 | when we were thinking about how to build artificial general intelligence, | 当我们想着如何建立人工一般情报时 |
| 524 | actually have it benefit all of humanity. | 实际上,它有益于全人类。 |
| 525 | How are you supposed to do that? | 你怎么能这么做? |
| 526 | And that the default plan of being like, | 而默认的计划是, |
| 527 | well, you build in secret, you get the super-powerful thing, | 好吧,你秘密建造, 你得到超强的东西, |
| 528 | and then you figure out the safety of it, | 然后你找出它的安全, |
| 529 | and then you push go and you hope you got it right. | 然后再推下去,希望你是对的 |
| 530 | I don't know how to execute that plan. | 我不知道如何执行那个计划。 |
| 531 | Maybe someone else does, but for me, that was always terrifying. | 也许别人知道 但对我来说 那总是很可怕 |
| 532 | It didn't feel right. | 感觉不对劲 |
| 533 | And so I think that this alternative approach | 所以我认为这个替代方法 |
| 534 | is the only sort of other path that I see, | 是唯一一种其他路径 我看见, |
| 535 | which is that you do let reality hit you in the face. | 那就是你让现实袭击你的脸。 |
| 536 | And I think you do give people time to give input. | 我认为你确实给人们一些时间提供投入。 |
| 537 | You do have what, before these machines are perfect, | 在这些机器完好之前 你确实有办法 |
| 538 | before they are super-powerful, | 在他们超能力之前, |
| 539 | that you actually have the ability to see them in action. | 你确实有能力看到他们行动起来 |
| 540 | And we've seen it from GPT-3. | 我们从GPT -3看到它。 |
| 541 | GPT-3, we really were afraid that the number one thing | GPT-3,我们真怕头号 |
| 542 | people were going to do with it was generate misinformation | 人们会为此制造错误信息 |
| 543 | and try to tip elections. | 并尝试给选举小费。 |
| 544 | Instead, the number one thing was generating Viagra spam. | 倒不如说,第一是产生威哥垃圾邮件。 |
| 545 | So, Viagra spam is bad, but there are things that are much worse. | 所以,伟哥的垃圾邮件是坏的, 但有些事情更糟糕。 |
| 546 | Here's a thought experiment for you. | 给你做个思想实验 |
| 547 | Suppose you're sitting in a room, there's a box on the table. | 假设你坐在房间里,桌子上有盒子 |
| 548 | You believe that in that box is something that there's a very strong chance | 你相信那个盒子里有很强的机会 |
| 549 | it's something absolutely glorious | 这是绝对光荣的东西 |
| 550 | that's going to give beautiful gifts to your family and to everyone. | 这会给你的家人和所有人 带来美好的礼物 |
| 551 | But there's actually also a one percent thing in the small print there | 但实际上还有百分之一的东西 在小的指纹那里 |
| 552 | that says, | 说, |
| 553 | (原始内容存档于2018-09-27). Pandora. | |
| 554 | And there's a chance that you're going to see a picture of Pandora | 你会看到潘多拉的照片 |
| 555 | and there's a chance that this actually could unleash | 并有一个机会 这实际上可能释放 |
| 556 | unimaginable evils on the world. | 无法想象的邪恶在世界上。 |
| 557 | Do you open that box? | 你打开那个盒子吗? |
| 558 | Well, so absolutely not. | 那么,绝对不是。 |
| 559 | I think you don't do it that way. | 我想你不会这么做的 |
| 560 | And honestly, I'll tell you a story that I haven't actually told before, | 老实说,我会告诉你一个我以前没有讲过的故事, |
| 561 | 我们开始OpenAI后 | |
| 562 | I remember I was in Puerto Rico for an AI conference. | 我记得我在波多黎各参加AI会议 |
| 563 | I was sitting in the hotel room, | 我坐在酒店房间 |
| 564 | just looking out over this wonderful water, | 只是看着这美妙的水, |
| 565 | all these people having a good time, | 这些人玩得很开心 |
| 566 | and you think about it for a moment. | 你考虑一下 |
| 567 | And if you could choose for a potentially, basically that Pandora's box, | 如果你可以选择一个潜在的, 基本上是潘多拉的盒子, |
| 568 | to be five years away or 500 years away, | 5年或500年之后, |
| 569 | which would you pick? | 你会选哪一个? |
| 570 | And on the one hand, you're like, | 一方面,你喜欢, |
| 571 | well, maybe for you personally, it's better to have it be five years away. | 好吧,也许对于你个人来说, 它最好有五年远。 |
| 572 | But if it gets to be 500 years away and people get more time to get it right, | 但是,如果它得到500年的远 和人们得到更多的时间 得到它的权利, |
| 573 | which do you pick? | 你选哪一个? |
| 574 | And I just really felt it in that moment. | 我当时真的感觉到了 |
| 575 | I was like, of course you do the 500 years. | 我当然喜欢你做500年。 |
| 576 | For real, my brother was in the military at the time, | 说真的,我哥哥当时是军人 |
| 577 | and he puts his life on the line in a much more real way | 他以更真实的方式 将他的生命置于现实之中 |
| 578 | than any of us typing things in computers | 比我们任何人都在电脑上打字 |
| 579 | and developing this technology at the time. | 和开发这一技术 在当时。 |
| 580 | And so, yeah, I'm really sold on the, you've got to approach this right. | 所以,是的,我真的很卖 这一点,你必须正视。 |
| 581 | But I don't think that's quite playing the field as it truly lies. | 但我不认为这是在玩 因为它真的谎言。 |
| 582 | If you look at the whole history of computing, | 如果你看看整个计算历史, |
| 583 | I really mean it when I say that this is a industry-wide, | 我是认真的 当我说这是一个全行业, |
| 584 | or even sort of just almost like a human development of technology-wide shift. | 甚至有点像整个技术转变的人类发展。 |
| 585 | And the more that you sort of don't put together the pieces that are there, | 你越是不把那些碎片拼凑在一起 |
| 586 | we're still making the faster computers, | 我们还在制造更快的电脑 |
| 587 | we're still improving the algorithms, | 我们还在改进算法 |
| 588 | like all these things, they are happening. | 像所有这些事情,它们正在发生。 |
| 589 | And if you don't put them together, you get an overhang. | 如果你不把它们拼凑起来 就会有麻烦了 |
| 590 | Which means that if someone does, | 也就是说,如果有人这样做, |
| 591 | or the moment that someone does manage to connect to the circuit, | 或者当有人设法连接到电路时, |
| 592 | then you suddenly have this very powerful thing, | 然后你突然有了这个非常强大的东西, |
| 593 | no one's had any time to adjust, | 没有人有时间调整, |
| 594 | like who knows what kind of safety precautions you get. | 就像谁知道你得到了什么样的安全防范措施。 |
| 595 | And so I think that one thing I take away is, | 所以我认为我拿走了一件事 |
| 596 | even if you think about development of other sort of technologies, | 即使你想着开发其他类型的技术 |
| 597 | think about nuclear weapons, | 想想核武器, |
| 598 | people talk about being like a zero-to-one sort of change | 人们在谈论一个零对一的变化 |
| 599 | in what humans could do, | 以人类所能做的事发誓, |
| 600 | but I actually think that if you look at capability, | 但我真的认为 如果你看看能力, |
| 601 | it's been quite smooth over time. | 好久不见 |
| 602 | And so the history, I think, of every technology we've developed | 所以,我想,我们开发的每一个技术的历史 |
| 603 | has been you've got to do it incrementally, | 一直以来,你必须逐步地做到这一点, |
| 604 | and you've got to figure out how to manage it | 你必须想出办法 |
| 605 | for each moment that you're sort of increasing it. | 每一瞬间你都在增加 |
| 606 | CAOKEY MCCLELLAN So what I'm hearing is that you, | 考基·梅克莱伦 所以,我听说的是,你, |
| 607 | the model you almost have is that we have birthed | 你几乎拥有的模型是 我们出生了 |
| 608 | this extraordinary child that may have superpowers | 这个非凡的孩子 可能拥有超能力 |
| 609 | that take humanity to a whole new place. | 将人类带到一个全新的地方 |
| 610 | It is our collective responsibility | 这是我们的集体责任 |
| 611 | to provide the guardrails for this child, | 为这个孩子提供护栏 |
| 612 | to collectively teach it to be wise and not to tear us all down. | 集体地教它要聪明,不要把我们都撕碎. |
| 613 | Is that basically the model? | 那基本上是模型吗? |
| 614 | CAOKEY MCCLELLAN I think it's true, | 考基·梅克莱伦 我想这是真的 |
| 615 | and I think it's also important to say this may shift, right? | 我想也有必要说这可能会改变 对吧? |
| 616 | We've got to take each step as we encounter it, | 我们必须在遇到它时采取每一步, |
| 617 | and I think it's incredibly important today | 我觉得今天很重要 |
| 618 | that we all do get literate in this technology, | 我们都在这种技术上识字, |
| 619 | figure out how to provide the feedback, decide what we want from it. | 找出如何提供反馈, 决定我们想要从它。 |
| 620 | And I think that my hope is that that will continue to be the best path, | 我认为,我希望,那将继续是最佳的道路, |
| 621 | but it's so good we're honestly having this debate, | 但很高兴我们真的有辩论 |
| 622 | because we wouldn't otherwise if it weren't out there. | 因为我们不会 否则,如果它不在那里。 |
| 623 | CAOKEY MCELLLLAN GREG Brockman, 非常感谢你来到TED。 | |
| 624 | and blowing our minds. | 吹动我们的思想。 |
| 625 | Thank you. | 谢谢 |
| 626 | CAOKEY MCCLELLAN Appreciate it. | 考基·梅克莱伦 谢谢 |
| 627 | CAOKEY MCCLELLAN Thank you. | 考基·梅克莱伦 谢谢 |
该视频共有字幕 627 条。解锁更多字幕为会员功能,请移动到 价格
