The AI arms race is on
发布时间 2023-02-13 23:54:58 来源
Okay, so I'm gonna try out chat GPT for the first time. This is one of our producers, Eliza Dennis. And last night, she decided to try out chat GPT.
好，那么我第一次尝试聊天GPT。这是我们的制片人艾丽莎·丹尼斯。昨晚，她决定尝试聊天GPT。
So I'm navigating to my web browser. Okay. Going to the open AI website. Yes, I am a human. Thanks for asking.
所以现在我要打开我的网页浏览器。好了。打开 Open AI 的网站。是的，我是一个人类。谢谢您的询问。
Chat GPT is a chatbot developed by a company called OpenAI. It uses generative artificial intelligence. The chatbot scrapes the internet to learn how to come up with an answer for pretty much whatever you ask it for. And it gets better the more it's used. When you talk to chat GPT, it can almost seem like you're talking to a human.
Chat GPT是由OpenAI公司开发的聊天机器人。它使用生成式人工智能。这个聊天机器人会在互联网上搜集资料，从而学会如何回答你提出的问题。而且，使用的次数越多，它就会变得越来越好。当你和Chat GPT聊天时，它几乎会让你感觉像是在和一个人类交流。
Okay, chat GPT. Let's see if you can make my job obsolete. Let's have you write an introduction with music for post reports podcast about chat GPT. Chat GPT has us starting with music. So hit it.
好的，与GPT来聊天吧。让我们看看你是否能让我失业。让你写一份有关聊天GPT的介绍，并加上音乐作为《邮报报告》播客的开场。聊天GPT要让我们从音乐开始。那就上吧！
Some music plays a slow, building electronic track with a pulsing beat. Welcome to the post reports. Music builds to a climax. Exploring the frontiers of technology where the future is shaped by the power of artificial intelligence. Music fades out to a quiet, steady pulse. And today, we delve into the world of chat GPT. Music continues to pulse in the background.
有些音乐播放着一个慢慢建立的电子曲目，带有脉动的节奏。欢迎来到这篇后续报告。音乐被逐渐强化到高潮。我们将探究技术的前沿，未来是由人工智能的力量所塑造的。音乐逐渐淡出，节奏保持平稳。今天，我们将深入聊天GPT的世界。音乐在背景中继续跳动着。
The language model created by OpenAI, capable of understanding and generating human-like text. This cutting-edge technology has the potential to revolutionize the way we interact with machines and open up new possibilities for AI applications. Music crescendos to a dramatic climax.
OpenAI创建的语言模型可以理解和生成类似人类的文本。这项尖端技术有潜力彻底改变我们与机器的互动方式，并为人工智能应用开辟新的可能性。音乐随着高潮渐增，达到了紧张的高潮。
Chat GPT doesn't usually talk out loud like this. It lies a audit to answer this way by using a text-to-speech function. But either way, we were kind of impressed by its writing chops. Here's Tech reporter Natasha Tiku.
“Chat GPT通常不会像这样大声说话。它通过使用文本转语音功能来回答问题。但无论如何，我们都对它的写作水平感到印象深刻。” 这是科技记者娜塔莎·蒂库说的话。
Chat GPT is a piece of generative AI that lets users ask questions, generate essays, spin up sign-feld scripts, and get kind of a conversational response from the chatbot. And it's an example of this new technology generative AI that's kind of taken the internet by storm. When Chat GPT was released to the public, people did have a lot of fun with it.
Chat GPT 是一种生成性人工智能，让用户可以提出问题、撰写文章、创建符号场脚本，并从聊天机器人中获得一种对话式的回应。它是这种新技术生成性人工智能的一个例子，现在非常流行于互联网上。当 Chat GPT 对公众发布时，人们非常喜欢它，觉得它很有趣。
People were asking just random things like write a biblical verse in the style of the King James Bible, explaining how to remove peanut butter from a VCR, and explain physics to me like I'm a five-year-old, or you could say, tell me the bill of rights in a limbic form. In most cases, it's able to generate these responses almost instantaneously, kind of depending on how overworked open AI servers are.
有些人正在随意地询问一些问题，比如用国王詹姆斯圣经的风格写下一节经文，解释如何将花生酱从录像机中清理出来，用五岁孩童的方式解释物理，或者你还可以说，用脑辅神经系统的方式告诉我宪法修正案。在大多数情况下，它几乎可以立即生成这些回答，这有点取决于 OpenAI 服务器的工作负载有多重。
Natasha, who has been covering AI for years, isn't as charmed by this chat pod as a lot of other people were.
纳塔莎这几年一直报道人工智能，对这个聊天机器人并不像许多其他人那样感到迷恋。
Oh, yeah. I mean, I always think like, you know, when people get marvel at what Chat GPT says, I'm like, well, it's just reflecting human ingenuity back at you, right? Yeah. Exactly. You know, I just feel like we've moved from like strip mining, you know, the environment. We're just like strip mining humans for, for their like intelligence and their creativity.
哦，是啊。我的意思是，我总是认为，你知道的，当人们对Chat GTP说的话感到惊奇时，我就像在告诉你，这只是在回应人类的智慧。是的。完全正确。你知道的，我只是感觉我们已经从像挖掘矿物一样挖掘环境的阶段转移过来了。我们正在像挖掘人类一样去挖掘他们的智慧和创造力。
From the Newsroom of the Washington Post, this is Post Reports. I'm Ella Hay-Ezzati, your human host. It's Monday, February 13. Today, we talk about how this technology getting into the hands of average consumers has kicked off an AI arms race and the dangers that could come with this new Silicon Valley War. Join us as we dive into the world of Chat GPT and discover its potential impact on our future. Music fades out to end the introduction.
这里是《华盛顿邮报》的新闻室，欢迎收听《邮报报道》。我是Ella Hay-Ezzati，您的人类主持人。今天是2月13日，我们要谈论的是，这种技术进入普通消费者的手中引发了一个人工智能竞争，并可能带来的危险。加入我们，探究Chat GPT世界，了解它对我们未来的潜在影响。音乐淡出，结束自我介绍。
So Natasha, last week, there were these big tech companies that announced how they're going to integrate chat bots like Chat GPT into some of their existing products, essentially starting what seems like an AI arms race. And I do want to get into the consequences of that later and what that all means. But first, can you just explain the technology behind Chat GPT and how it's different than other predictive texts that we might encounter before with, I don't know, like customer service bots and that sort of thing?
嗨 Natasha，上周有一些大型科技公司宣布了他们将如何将像Chat GPT这样的聊天机器人集成到他们现有的产品中，本质上开启了一场人工智能竞赛。我想稍后会讨论这一切的后果以及意义。但首先，你能否解释一下Chat GPT背后的技术，以及它与我们以前遇到的其他预测文本（例如客户服务机器人等）有何不同？
Yeah, it's much more sophisticated than what you would have previously encountered. I think we all know how stilted and awkward and really unhelpful those customer service bots can be. These models in contrast have been trained on massive amounts of data, text, scrape from the web so you can think like a corpus of books, links from Reddit and they have been told to essentially predict the next word in a sentence. And because this is machine learning, the model actually kind of teaches itself how to find patterns between words. And the developers have found that these models were able to generate really human-like, kind of instantaneous original responses to questions. So the end result is like a much more fluid back and forth. And something that people found really appealing, it just was leaps and bounds from like a corporate chatbot.
嗯，它比你以前遇到的要高级得多。我想我们都知道客户服务机器人是多么生硬、尴尬和真正没有用。相比之下，这些模型经过了大量数据的训练，包括文本和从网上抓取的信息，就像一些书籍的语料库、Reddit的链接等，它们已经被告知要预测句子中的下一个单词。由于这是机器学习，模型实际上会自己找到单词之间的模式。并且开发人员发现这些模型能够生成类似于人的、快速而原始的回答问题的方式。所以最终结果就是更加流畅的来回交流。人们发现这些模型非常吸引人，离企业聊天机器人还有很大的差距。
What are some of the ways this technology can be used? Like what are researchers saying about the need and benefit for this technology and why we even need it?
这项技术可以用在哪些方面？研究人员对于这项技术的需求和好处有什么看法，为什么我们需要它？能详细说一下吗？
Well, one thing people should know is that these models are already in use. You just normally don't see them. You don't encounter them face to face.. They're usually more at the infrastructure level.
嗯，人们应该知道的一件事是，这些模型已经在使用中。你通常看不到它们，也不会面对面遇到它们。它们通常更多地在基础设施层面上运作。
So Google, auto complete in your emails, content moderation on Facebook, language translation via Google, all of this relies on these same models. But this was the first time for almost all of us that we had access to this state of the art technology that is normally locked up behind closed doors in corporate labs. Like Google, like Facebook, they keep it under wraps.
所以，谷歌在您的电子邮件中自动完成，Facebook上的内容审核，通过谷歌的语言翻译，所有这些都依赖于相同的模型。但几乎所有人都是第一次接触到这种尖端技术，这种技术通常被锁在公司实验室的闭门之后。就像谷歌和Facebook一样，他们将其保密。
And when they do release it, it's in a really neutered form. And here OpenAI had put this out as a consumer product for free. For anyone to use, you could access it just from your web browser. And all of a sudden, you are able to play with this state of the art technology.
当它们发布它的时候，它的形式被削弱了许多。而OpenAI已经把这个作为一个免费的消费产品发布出来。任何人都可以使用它，只需从你的网络浏览器中访问即可。突然间，你就能够玩耍这项最新技术了。
You can almost think about it like Wikipedia on demand. You know how Wikipedia is really good at kind of summarizing really complex concepts. You can ask it to explain things at different levels, like two of fifth graders, two of high school students. It has a wide breath of knowledge. Which chances are if it was on the internet a lot, chat GVT knows it.
你可以想象它就像按需使用的维基百科一样。你知道维基百科在概括非常复杂的概念方面非常出色。你可以要求它用不同层次的方式解释事物，比如小学五年级的水平和高中生的水平。它具备广泛的知识面，如果它在互联网上很受欢迎，聊天机器人 GVT 就很有可能知道这个知识点。
People use it as a way to generate ideas. It's really successful. I think that way, like say you're having writers block or something like that. So it's good as like a creative tool in that way. I mean, listen, like if it were accurate, I would love this because in theory, it is like the best little assistant you had.
人们使用它来产生创意。它真的很成功。我认为这样做就像你遇到写作障碍或其他困难时的方式。所以从那个角度来说，它是一种很好的创意工具。我是说，听着，如果它是准确的，我会很喜欢它，因为从理论上讲，它就像你拥有的最好小助手。
So the pontification about how this could change the internet is totally unbridled. There's a lot of enthusiasm here in Silicon Valley. So it's being called the next platform. There's been a lot of anticipation of, okay, after mobile, after the cloud, like what's going to be the next big thing? And these models are really being heralded as the next mode of the web.
那些宣称这将改变互联网的说法，有点过分了。硅谷很兴奋。他们称它为下一个平台。我们一直期待着，在移动、云端之后，下一个大事情会是什么？这些模型被誉为网络的下一个模式。
We saw that with Dolly too, which was a text to image generator. It became really popular last summer.
我们也看到了多莉，它是一个文本到图像的生成器。它在去年夏天变得非常流行。
These generative models that make it easy for people to create music, text, essays, you know, cheat on your college exam will really transform the way that we interact online.
这些生成模型极大地方便了人们创造音乐、文本、论文，甚至在大学考试作弊，这将真正改变我们在线交流的方式。
Natasha, can you also give us an overview of what are some of the potential dangers of this technology and especially releasing it to the public this way?
娜塔莎，你能否给我们概述一下这项技术可能存在哪些潜在危险，特别是以这种方式向公众发布？如果有需要，可以改写。
Yeah, I think, you know, the biggest danger is really that we don't know exactly what's going to happen. This is really untested, you know, in a lot of ways. It's going from the lab to billions of people without that much vetting.
嗯，我觉得，你知道的，我们最大的危险是我们不确切地知道会发生什么。在很多方面，这真的是未经测试的。它从实验室走向了亿万人口，但并没有经过太多验证。
And if you look at the people who built this technology, it's very homogenous. It's oftentimes like very white, very male, very Asian. In addition to that, you know, these models really reflect the data that they were trained on.
如果你观察一下建立这项技术的人，就会发现他们非常同质化，通常是非常白、非常男性、非常亚洲的人。此外，这些模型真正反映了它们所训练数据的特点。
That data was scraped from a narrow portion of the web. It's the English speaking portion of the Western web for the most part. And I think if you've ever been online, if you've ever been on Reddit, you understand that that is probably going to come with a lot of bias and stereotypes.
这些数据是从互联网的一小部分收集而来。它主要来自英语为母语的西方互联网。如果你曾经上过网，如果你曾经在 Reddit 上，你就会明白这很可能带来很多偏见和成见。
And that is reflected in the types of text and behaviors that the model generates. So, that's something where the developers behind these models, such as OpenAI, they have tried to fix that by cleaning up the data sets to make the machine less racist, spit out less hate speech, less bias towards women. You know, they do take out some of the porn and gore and violence from the data sets.
这反映在模型产生的文本和行为类型上。因此，这是这些模型背后的开发人员，例如OpenAI等人，试图通过清理数据集来减少机器的种族歧视、仇恨言论和对女性的偏见。你知道，他们会从数据集中剔除一些色情、血腥和暴力内容。
And then after the training process, they put filters on what the machine is allowed to generate. But it's still being put out there without specific end goal in mind. So they're kind of just waiting to see how people use it.
然后在培训过程之后，他们为机器添加了过滤器以限制其产生的内容。但是这些内容还是被公开呈现且没有特定的目标。所以他们现在只是等待看看人们将如何使用它。
And when you do that with billions of people in a product that is super influential online, you're willing to take that risk. And I think for technological historians, for AI ethicists, for people who have released products before, that's just not the way that it's supposed to go.
当你在一个拥有数十亿用户，并且具有超级在线影响力的产品中这样做时，你愿意承担这样的风险。我认为对于技术历史学家、AI伦理学家和之前发布产品的人来说，这并不是应该遵循的方式。
Developers of these kinds of tools have been pretty clear that these tools have a real tendency to give inaccurate information. And not only do they give inaccurate information, but they also do something called hallucinate, which is when they give inaccurate information confidently.
这类工具的开发者已经非常明确地指出，这些工具很容易提供不准确的信息。而且它们不仅提供不准确的信息，还存在所谓的“幻觉”，也就是它们自信地提供不准确的信息。
And it's not a problem that the industry has solved. They've made some progress towards it, but they're releasing it to the public. So you have this system that is being treated as this like all-knowing, super useful tool.. It's being marketed to people as the future of information, the future of the internet, and it might be wrong, confidently wrong, and you won't know when. You know, and all you get is maybe a little bit of warning when you log on and a little disclaimer at the bottom.
这个问题行业还没有解决。虽然他们已经在朝着这个方向取得了一些进展，但他们正在向公众发布它。所以你有一个被视为全部知晓、超级有用的工具的系统。它被营销给人们作为信息和互联网的未来，但它可能是错误的，有自信地错误，而你不会知道。你知道，在你登录时，你可能只会得到一点点警告，还有一个小免责声明在页面底部。
Because we know from, I don't know, the last 10 years, especially, there is a lot of debate about truth online. And here we are introducing a AI model in all sorts of places online that is confidently giving us wrong information. And there's not a lot of AI literacy among online readers about how to interpret this. Should we be looking at chat GBT like an Oracle? Should we trust it more than Google? Less than Google? Is it our friend? We don't really know how that's going to play out, but it can't help our current information dystopia.
因为我们从过去的10年中特别清楚，网上的真相问题已经引起了很多争议。在这种情况下，我们引入了一种AI模型用于各种网络场合，但它却自信地给我们提供错误信息。而在线读者在如何解读这些信息方面缺乏足够的AI素养。我们应该像看待神谕一样看待Chat GBT吗？我们应该比信任Google更信任它吗？还是比Google更少信任它？它是我们的朋友吗？我们真的不知道未来会怎样，但它肯定不能帮助我们解决当前的信息恶化问题。
Natasha, tell me more about the company that developed chat GBT.
娜塔莎，请再向我介绍一下开发聊天GBT的那家公司。
Sure, the company behind chat GBT is super fascinating. It's called OpenAI. It was founded back in 2015 as a nonprofit actually by Elon Musk, Peter Teal and Sam Altman with a pledge to donate a billion dollars. And the idea was very different from the way the company operates now. The initial idea was that they wanted to provide an alternative to having super intelligence, you know, a really powerful AI be in the hands of a corporation like Google or a foreign government. So the idea was to make it open and transparent and distribute the benefits of AI to everyone around the world.
当然，背后推动聊天GBT的公司——OpenAI非常有趣。它是由埃隆·马斯克、皮特·蒂尔和萨姆·奥尔特曼于2015年成立的非盈利组织，承诺捐赠10亿美元。当时的想法与公司现在的运作方式非常不同。最初的想法是，他们想提供一个替代方案，即将超级智能（如强大的人工智能）掌握在谷歌等公司或外国政府的手中。所以，他们的想法是让它开放透明，并将人工智能的利益分配给全世界的人。
And I think they very quickly found that they weren't able to get enough money and like get further investment. So they turned from open to one of the most secretive companies working in AI. But they made this very big bet that the future of AI was going to be in making these models bigger and bigger. And it has really paid off. You know, they are now leading the race and they helped instigate a race actually towards putting out these models, these larger and larger models. This happens to be exactly what they said they didn't want. Their end goal is something called AGI. That stands for artificial general intelligence.
我认为他们很快意识到自己无法获得足够的资金和进一步的投资，所以他们从公开变成了AI领域最为神秘的公司之一。但是他们做出了一个非常大的赌注，也就是未来的AI会在于使这些模型变得越来越大。而这个赌注确实得到了回报，现在他们正在领先竞赛，他们实际上帮助发起了一场竞赛来发布这些更大的模型。这恰好是他们所说的不想要的。他们的最终目标是一种被称为AGI的人工智能，也就是人工通用智能。
It's this idea of AI that's comparable to human intelligence. You know, you can almost think about it as like the singularity. You know, it's this end goal that artificial intelligence experts have always been working towards or like some portion of the industry. It's very like sci-fi type goal, right? That was one just fringe, but that has become much more mainstream. So it's really now leading the pack and it has a relationship with Microsoft. And recently they just put a reported $10 billion into open AI.
嗯，就是关于AI的这种概念，它可以与人类智能相媲美。你知道，你几乎可以想象它就像是奇点。你知道，这是人工智能专家一直在追求的最终目标或行业的某个部分。它非常像科幻类型的目标，对吧？以前它只是一个边缘的想法，但现在它已经变得越来越主流了。所以它现在领导着这个行业，并且与微软有关系。最近他们刚刚投入了一笔大约100亿美元的资金到Open AI中。
So open AI focuses on putting out these products on the way to building AGI and then Microsoft is supposed to commercialize them. And the reason a lot more people are learning about this company and its name is becoming like much more of a household name is because of the way that their philosophy is around risks. They think that there's no way to like kind of fully make an AI model safe. So they need to release it into the public and see how people interact with it and closely monitor it. And that's the way you are going to kind of figure out where the dangers are.
所以，开放AI致力于通过打造AGI的过程来推出这些产品，然后微软负责商业化它们。这家公司的名字越来越为人熟知的原因在于他们的风险哲学。他们认为没有办法完全保证AI模型的安全性。因此，他们需要将其发布到公众中，观察人们与之交互的情况，并密切监测。这是你能够找出危险所在的方式。
After the break, Natasha and I talk about why big tech has been cautious for years about this technology. And why now they're throwing caution to the way. We'll be right back.
休息之后，纳塔莎和我将会谈论为何大科技公司多年来对这项技术持谨慎态度，以及为何现在他们不再谨慎。我们马上回来。
So Microsoft who has heavily invested in open AI made a surprise announcement last week. Natasha, what happened?
上周，Microsoft这家投资于开源人工智能的公司，做出了一个出人意料的声明。Natasha，请问发生了什么？
Yeah, it was, you know, maybe the worst kept surprise.
是的，你知道的，可能是最没有保密的秘密。
It's great to be here with all of you today. You've been working on something we think is pretty special.
今天与你们所有人在一起感觉很棒。你们一直在做一些我们认为非常特别的事情。
Microsoft announced that it would be incorporating a more updated version of chat GPT into its web browser Bing. Normally the punchline of a joke, right? That you probably haven't been to in a decade or maybe a decade and a half. And that it would be kind of radically altering its search offering.
微软宣布将在其网络浏览器必应中加入更新版本的聊天GPT。通常，这是一个笑话的结论，对吧？那么你可能已经十年或十五年没用过了。而且这将极大地改变其搜索服务。
Infused with AI and assembled as an integrated experience, we're going to reimagine the search engine, the web browser and new chat experiences into something we think of as your co-pilot for the web.
我们将加入AI技术并将其组装成一种整体体验，重新设想搜索引擎、网络浏览器和新型聊天体验，让它们成为我们为您设计的网络共驾员。
In order to compete with Google. With our innovation, they will definitely want to come out and show that they can dance. And I want people to know that we made them dance and I think that'll be a great day.
为了与谷歌竞争。凭借我们的创新，他们肯定想要出来表演一下自己的舞步。我希望人们知道是我们让他们跳舞，这将是一个伟大的日子。
Satya Nandela, the CEO of Microsoft, gave this quote to the verge and made it very clear that they are trying to get a bigger portion of this massive search market where Google has dominated for 20 years, 25 years.
微软的CEO Satya Nandela在接受The Verge采访时说了这样一句话，他非常明确地表示他们正在试图在谷歌已经占据了20年、25年主导地位的庞大搜索市场中争取更大的份额。
So can you describe Natasha how that will work? What's beneficial about using AI in a search engine and what's going to happen with Bing?
那么，你能描述一下对于纳塔莎，这个AI如何运作吗？在搜索引擎中使用AI有什么优点，对于Bing又会发生什么呢？
So the big change, the one that everyone was waiting for, is the way that you are able to interact with search. At the center of this new co-pilot experience is an all new Bing search engine and edge web browser. Not only does it give you the search results, but it will actually answer your questions.
所以，大家一直在等待的重大变化就是您与搜索互动的方式。这项全新的联合飞行员体验的中心是全新的Bing搜索引擎和Edge网页浏览器。它不仅会给您搜索结果，而且实际上会回答您的问题。
So they say you're able to put in like much more conversational queries. You know, there's no like Nike plus sneakers plus size five. We're going to let you chat. We're going to let you just talk to it naturally. You can just put in a question the same way you would talk to your friends.
所以，据说你能够输入更多像是日常会话的问题。你知道，没有类似于尺码为5的耐克加号鞋子之类的东西。我们会让你聊天。我们会让你自然地与它对话。你可以像与朋友聊天一样提出问题。
And the results come up like the same kind of links you'll see on the left side and on the right side, you'll get a chat GPT-like response. So a kind of summary written like a conversation and you'll see little kind of footnotes of where the information was sourced from. And then you'll also have an option of chatting back and forth and asking follow-up questions.
结果会显示与左侧或右侧的链接类型相同的链接，你将获得类似聊天的GPT响应。所以，这是一种按对话方式书写的总结，你将看到从哪里获得信息的小脚注。然后，你还可以选择来回聊天并提出跟进问题。
Say you said where are the best sneakers in size 6.5. And then you could ask, you know, where can I get them in blue? And then it has this other feature, the one that I'm really excited about, where you can open this up on any page on the web in the demo what they showed is opening up this feature on GAPS financial page. And they were able to get like a summary of the results and then compare it with the financial results for Lulu Lemon, like instantaneously. So it could let you say summarize an article as you're reading it or do any of the things you're able to do with chat GPT.
你的话说，你问最好的6.5尺码运动鞋在哪里可以买到。然后你可以问，你知道在哪里可以买到蓝色的鞋子吗？ 还有一项功能，这是我特别兴奋的，你可以在网页的任何页面打开这个功能，演示中显示的是在GAPS金融页面上打开这个功能。他们能够获得结果的摘要，然后立即将其与Lulu Lemon的财务结果进行比较。因此，它可以让你在阅读文章时进行摘要，或者像聊天GPT一样做任何事情。
And yet, you know, for Bing, I think this is just a total game changer for them. It's probably a once in a lifetime chance to try to catch up with Google, which has had somewhere between 80 and 90 percent market share forever.
然而，你知道的，对于必应来说，我认为这对他们来说是一个完全的游戏改变者。这可能是一生中一次机会，试图追赶谷歌，谷歌一直占据着80%到90%的市场份额。
We've long been pioneers in this space, not just in our research, but also in how we bring those breakthroughs to the world and our products in a responsible way. And then the very next day after this announcement from Microsoft, Google had their own, they demonstrated their own chatbot.
我们在这个领域一直是先驱者，不仅在研究方面，还在如何以负责任的方式将这些突破带给世界和我们的产品方面。然后就在微软宣布之后的第二天，谷歌也推出了自己的聊天机器人，并进行了演示。
Back at IO in 2021, we unveiled our Lambda AI models, a breakthrough in conversational technology. Next, we're bringing Lambda to an experimental conversational AI service, which we fondly call BARD.
2021年时，我们在IO大会上推出了我们的Lambda AI模型，这是对话技术的重大突破。接下来，我们将Lambda引入到一项实验性的对话AI服务中，我们亲切地称其为BARD。
So Natasha, was there anything noteworthy about that or that stood out to you there?
嗨，娜塔莎。在那里，有什么值得注意的事情或者对你来说特别突出的吗？
Well, I should say neither of these are fully accessible to the public yet. You can kind of get a little bit of a demo of Bing, which they're calling new Bing. The same your grandmother's Bing. Yes, but you can sign up for a wait list and the demo at Microsoft was an actual demo. You know, people could interact with it. Our tech columnist, Jeff Valor, was able to do a bunch of searches. In contrast, Google was showing what most people said was kind of just a design mockup. You know, they're saying that only some testers are getting access to this. So there's a big difference between like this is what it might look like and here, you know, we're shipping a product. We'll be able to use it soon here. A bunch of people are already trying it out.
嗯，我应该说这两个都还没有完全对公众开放。你可以尝试一下称之为新Bing的演示版。不同于老式的Bing。是的，但你可以在Microsoft注册等待列表并进行演示。这是一个真正的演示，人们可以和它互动。我们的技术专栏作家Jeff Valor能够进行了一系列搜索。相比之下，Google展示的是大多数人说是一种设计模型。他们说只有一些测试人员才能获得访问权。因此，在“这是它可能看起来像什么”和“我们正在发布一款产品”之间存在很大差异。很多人已经开始尝试了。我们很快就能用上它了。
So Google's kind of hastily put together answer to this was really not well received. Google's new highly touted AI chatbot BARD has already made a booboo.
谷歌匆忙地推出的答案并没有得到良好的反响。谷歌新推出的备受瞩目的AI聊天机器人BARD已经犯了一个错误。
Introduce this week, BARD was touted in an online ad by Google that ran in the company's Twitter feed. In the ad, BARD has given the prompt, quote, what new discoveries from the James Webb Space Telescope or JWST? Can I tell my nine year old about it mentioned that the James Webb telescope was the first time telescope was able to take a picture of an exoplanet that wasn't true. Oh, no.
本周，谷歌在公司的 Twitter 账户上发布了一则在线广告，宣传了 BARD。在广告中，BARD 发出提示：“詹姆斯·韦伯太空望远镜有哪些新发现？我能告诉我九岁的孩子吗？”然而，该广告中提到詹姆斯·韦伯望远镜是第一次拍摄到一颗外行星的照片，这是不正确的。哦，不。
The error was spotted hours before Google hosted a launch event for BARD in Paris where Google senior executive touted BARD as the future of the company. And the internet just realized this at the same time that they're doing this demonstration and the stock ended up tanking $100 billion.
在巴黎举行的BARD发布会前数小时就发现了错误，当时Google高级主管还把BARD吹嘘为公司的未来。然而，正当他们正在展示时，互联网也发现了这个错误，结果导致股价暴跌1000亿美元。
Yes. It's just like Wall Street, twitchy fingers and the amount of hype and obsession and interest around generative AI right now. This is not based on like a fully thought out understanding of how this all might shake out.
是的。就像华尔街一样，现在对生成式人工智能的狂热和痴迷以及关注度很高，手指不停地动来动去。这并不是基于对这一切可能会如何发展的深思熟虑的理解。
Google, it should be said, is the original developer behind a lot of the core components of generative AI. It is the place where transformers, this architecture that is used to build these models was first developed. It has its own language models. You know, it does not use them as consumer tools. It's been really late to the game, but yeah, it was, you know, almost makes you feel bound for a truly not a company. Almost.
Google是生成式AI许多核心组件的原始开发者。它是transformer的诞生地，这种架构是用来构建这些模型的。它有自己的语言模型。它不会将其作为消费者工具使用。它真的很晚才进入游戏，但是是的，它几乎让你觉得它不是真正的一个公司。几乎。
And what about companies outside of Microsoft and Google? Is anyone else in this sort of arms race to sharpen their products with AI?
还有微软和谷歌以外的公司呢？还有其他公司在这种通过人工智能来提升他们的产品的竞争中吗？
Oh, my God. I saw this map of generative AI startups. Like my eyes just widened.
哦，我的天啊。我看到了这张生成型人工智能创企地图，我的眼睛都快瞪出来了。
Yes, there's a ton of money flowing into generative AI startups. There is like a bottleneck because in order to build these models, you have at least up until now needed a lot of access to data and needed a lot of money for compute power.
嗯，很多钱正在涌入生成式人工智能初创公司。这就像一个瓶颈，因为要建立这些模型，至少到目前为止需要大量获取数据，并需要大量的计算能力支持。
And so only very few labs like DeepMind, which is owned by Google, Google, OpenAI, a company called Anthropic, you know, companies that have raised hundreds of millions or billions of dollars have been able to compete. But some of them are able to access OpenAI's software, which they're making available to businesses. Google has said it will make its large language model available to businesses. So this is definitely not the end that you will hear about it.
所以只有像 DeepMind 这样由 Google 拥有的实验室、Google、OpenAI、一家名叫 Anthropic 的公司等仅有一些已经筹集了数亿或数十亿美元资金的公司能够竞争。但是其中一些公司能够获得 OpenAI 开发的软件，这些软件现在已经对商业开放。Google 表示将使其大型语言模型可用于商业用途。因此，这肯定不是我们听到的最后一个消息。
So I mean, clearly because there's so much investment in this space, it must mean that this technology companies are viewing it as being very valuable and eventually very profitable.
我是说，显然因为这个领域有如此多的投资，这一定意味着科技公司认为它非常有价值，最终非常赚钱。
So Natasha, can you just explain or break down like how is it that generative AI is going to make money for companies? Is it just like, oh, this technology will replace human workers? Are my being too cynical about that?
嗨，娜塔莎，你能解释一下产生式人工智能是如何为公司创造收益的吗？是像这样，这项技术会取代人类工人吗？我的想法太愤世嫉俗了吗？
I mean, I don't think you're being too cynical because that has been the mantra we've been hearing from Silicon Valley. I mean, it's also kind of a trope when they make a big breakthrough. They like to say like who it's going to put out of business.
我的意思是，我不认为你太愤世嫉俗，因为这一直是我们从硅谷听到的口号。我的意思是，当他们取得重大突破时，他们也喜欢说会让谁失业，这也有点老套了。
I always think about the self-checkout at drug stores or at grocery stores. And this is from listening to really smart historians about the history of technology. Very often they talk about automation as eliminating the need for humans, which it certainly has gotten rid of entire categories of jobs. But oftentimes what happens is that sense that technology is coming in that fear allows employers to drive down wages and make the job more perilous. But the human are still needed to be there because the automation, the machine, is not flawless.
我总是想到药店或杂货店的自助结账。这是因为我听了非常聪明的历史学家讲述技术的历史。他们经常谈到自动化消除了人类的需要，它确实淘汰了整个职业类别。但通常情况下，技术即将到来的那种恐惧感使雇主可以降低工资并使工作更加危险。但人类仍然需要在那里，因为自动化、机器并不完美。
And it still needs help and it's not perfect. And you need to be there to check it. So here you are kind of like working alongside it. At this particular moment in time, you also have this other kind of countervailing force of people in Silicon Valley who really believe in this technology and they want to see it released in this unbridled way. They call themselves like accelerationists. And they're kind of happy to see some of these job categories fall away. So there's still just so much to be determined.
它仍需要帮助，而且并不完美。而你需要在那里去检查它。因此在某种程度上，你是在它旁边工作。在此特定时刻，你还有另一种反制力量来自硅谷的人们，他们真的相信这项技术，希望以无限制的方式推出它。他们自称为加速主义者。他们对这些工作类别的消失感到高兴。所以还有很多事情有待确定。
It's still such an early stage and the models are changing so fast. And certainly there have been a lot of job categories that they say this will put out of business. Writers, marketers, artists. So there is the idea that businesses will pay for this software rather than pay for advertising companies or marketing companies, pay for this rather than writers. But I'm sure you will also be seeing a number of startups that just can't make the numbers work in their favor. But if there's enough venture capital flowing their way, they can be venture subsidized for a while.
现在阶段还很早，模型变化非常快。当然，已经有许多行业被认为将被淘汰，比如写作者、营销人员、艺术家等。因此，有一种想法是企业会为这个软件付费而不是为广告公司或营销公司付费，也不需要为作家付费。但我相信你也会看到许多创业公司无法使数字对他们有利。但如果有足够的风险投资流向他们，他们可以被风投资助一段时间。
So Natasha, we're seeing this AI arms race really heating up in Silicon Valley and given everything we've discussed, should we be worried like what are the consequences of these companies racing into this technology?
那么，纳塔莎，我们看到这种 AI 武器竞赛在硅谷真的热闹起来了，考虑到我们所讨论的一切，我们应该担心吗？这些公司竞相涉足这项技术会有什么后果呢？
I'm worried. I was talking to an AI engineer recently who said it just feels like everyone is waiting with baited breath. It's really hard to tell exactly how this will be received, especially when it's going from straight from the lab into the hands of billions of people through search, which is like one of the most impactful parts of the web, right? It's how you get your information. It's like your portal to knowledge for most of us, for me at least.
我很担心。最近我和一个AI工程师交谈时，他说似乎每个人都在屏息期待。很难准确地判断这会被接受到什么程度，特别是当它从实验室直接走进亿万人手中，通过搜索引擎，这是网络中最有影响力的部分之一，对吧？它是获取信息的方式。对于我们大多数人来说，它就像是知识之门，至少对我来说是这样。
And AI ethicists, technologists, researchers themselves have been warning about this race dynamic for a long time because the arms race analogy is apt, right? They argue that someone else is going to be doing it. So we have to do it and it incentivizes companies or it gives companies a justification for putting technology out there without fully testing the potential harms and risks without thinking about safety.
人工智能伦理学家、技术专家和研究人员自己已经长期警告这种赛跑动态，因为赛跑的类比是恰当的，对吧？他们认为，别人会去做这个。所以我们必须做，这会激励公司，或者给公司以推出技术而不完全测试潜在危害和风险，而不考虑安全的理由。
I mean, we've seen how this goes, right? Like we've seen an arms race before. We actually just lived through the results of Silicon Valley's last arms race. At the time, we called it growth at all costs or if you're familiar with the Facebook motto, move fast, break things. And we know how that ended for democracy, for civic values, for the way that we ingest information, for political polarization, for our ability to, I don't know, talk to our neighbors.
听着，我们都知道这个结局会怎样，对吧？就像我们以前见过的那场军备竞赛。我们刚刚经历了硅谷上次武装竞赛的结果。当时，我们称之为不惜代价的增长，或者如果你熟悉 Facebook 的座右铭，就是快速行动，打破一切。我们知道这对于民主、公民价值观、我们摄取信息的方式、政治极化、以及我们与邻居交流的能力，都带来了什么样的后果。
So just as the public is kind of able to give feedback to these companies that we want you to take our safety seriously, we want you to take your impact in the real world seriously and have some accountability.
所以，公众可以向这些公司提供反馈，表达我们希望你们认真对待我们的安全，认真对待对现实世界的影响，并承担一定的责任。
Here we have this arms race for AI, which is completely emblematic of the same problems we saw through social networks, which is like that the developers don't often know why the technology is doing what it's doing.. You know, it might not know why it is giving one answer and not the other.
我们现在正在进行AI的武器竞赛，这完全象征着我们在社交网络中遇到过的同样问题，即开发人员经常不知道技术为什么会做出某些决策。你知道，AI可能没有清楚为什么会给出某个答案而不是其他答案。
It was instructed to just find patterns between words and try to please you when you ask it a question. So yeah, we're back to move fast, break things and we know how that went the last time.
我们的指示是仅查找单词之间的模式，并在您提出问题时尝试取悦您。所以，是的，我们又回到了快速前进、打破陈规的模式，我们知道上一次会发生什么。
Natasha, thank you so much for your time. Thanks for having me. Natasha Tiku covers Silicon Valley for the post. Special thanks to Rachel Lerman for her reporting on Microsoft.
娜塔莎，非常感谢你花时间和我交流。谢谢你的邀请。娜塔莎·提库为《华盛顿邮报》报道硅谷的新闻。还要特别感谢瑞秋·莱曼为我们报道了微软的新闻。
That's it for post reports. Thanks for listening. Today's show was produced mixed and edited by humans, namely Eliza Dennis, Sam Bear and Maggie Pemman.
这就是本期的报道内容。感谢您的收听。今天的节目由人类制作、混合和编辑，他们分别是艾丽莎·丹尼斯、萨姆·贝尔和玛吉·佩曼。
I'm Ella Hay Ezzati. We'll be back tomorrow with more stories from the Washington Post.
我叫Ella Hay Ezzati。明天我们将回来，为大家带来更多的《华盛顿邮报》的故事。