今年AI领域的精华荟萃

我们今天这一集将以一个有点老套的方式开场——一个字典定义。

We are going to kick off today's episode with a little bit of a cliche, a dictionary definition.

但请继续关注我们。

But stick with us.

这很相关。

It's relevant.

而《梅里亚姆-韦伯斯特词典》迫使我们不得不这么做，因为他们将‘slop’选为了年度词汇。

And Merriam Webster forced our hand by making slop its word of the gear.

他们的定义是：由人工智能大量生成的低质量数字内容。

Their definition, digital content of low quality that is usually produced in quantity by means of artificial intelligence.

你懂的。

You know it.

你见过的。

You've seen it.

这很奇怪。

It's weird.

它笨拙不堪，而且无处不在。

It's clunky, and it is everywhere.

Vuco la carte。

Vuco la carte.

等等，孩子。

Wait, son.

让我

Let me

给我一些。

have some.

爸爸，看看这个煎饼飞艇。

Dad, check this pancake blimp.

真糟。

Nuts.

他在那儿，警官。

There he is, officer.

他偷了我的煎饼。

He stole my pancakes.

YouTube频道Funtastic YT一直在发布以卡通猫为主题的短视频。

The YouTube channel Funtastic YT has been cranking out short videos featuring cartoon cats.

这些视频已经获得了数亿次观看。

It's racked up hundreds of millions of views.

一家主要报纸发布了一份由人工智能生成的夏季阅读清单，其中包含了10个虚构的书名，这些书名却署名了真实畅销书作者，比如安迪·威尔的《最后的算法》或珀西瓦尔·埃弗雷特的《降雨者》。

An AI generated summer reading list was published by major newspapers and featured 10 made up titles from real best selling authors, like The Last Algorithm by Andy Weir or The Rainmakers by Percival Everett.

播放列表也被误导了。

Playlists were also slumped.

今年夏天，一支名为Velvet Sundown的乐队的音乐充斥了Spotify用户的“发现周刊”推荐流。

This summer, music from a band called the Velvet Sundown flooded Spotify users discover weekly feeds.

在人们发现这支乐队根本不存在之前，它的音乐已经被播放了数百万次。

And within a few weeks, the band's music had been streamed millions of times before anybody noticed the group wasn't actually real.

挥舞着黑白旗帜。

Waving flags in black and white.

九月，Spotify宣布已从其平台移除了七千五百万条垃圾音轨。

In September, Spotify announced it had removed 75,000,000 spammy tracks from its platform.

但其他公司却开始沉溺于这些垃圾内容中。

But other companies began to wallow in the slop.

就在同一周，Meta推出了一个供用户创建和分享AI生成视频的动态消息流。

In the same week, Meta released a feed for users to create and share AI generated videos.

OpenAI也推出了一款用于生成视频和音频的新应用版本。

And OpenAI launched a new version of an app to generate video and audio.

就连全球最大的唱片公司之一华纳音乐集团，上个月也与两家曾因侵犯版权而起诉它的AI公司达成了授权协议，允许其平台的付费用户使用那些同意参与的艺术家的声音和作品创作歌曲。

Even one of the world's biggest record labels, Warner Music Group, signed a licensing deal last month with two AI companies that was previously suing for copyright infringement, allowing paid users of their platforms to create songs with the voices and compositions of artists who agreed to participate.

如今，AI垃圾内容甚至渗透进了该国最古老的词典中。

And now AI Slop has managed to find its way into the pages of the country's oldest dictionary.

请考虑一下。

Consider this.

今年已经证明，AI正在迅速重塑网络现实，而这些垃圾内容将长期存在。

The year has proved that AI is rapidly reshaping online reality, and the slop is here to stay.

接下来，我们将回顾2025年最出色的垃圾作品，并教你如何分辨真假。

Coming up, we will take a look at the cream of the slop from 2025 and how to sort the real from the fake.

来自NPR，我是斯科特·德特罗。

From NPR, I'm Scott Dettrow.

在迎接2025年跨年倒计时时，不妨收听NPR音乐频道的《All Songs Considered》播客，一起回顾今年最重磅的歌曲与专辑。

As you prepare for the ball drop on 2025, listen to NPR Music's All Songs Considered podcast as we look back at the biggest songs and albums of the year.

从不容错过的热门单曲，到你可能还没听过的精彩冷门作品。

From the unmissable hits to the fascinating other stuff you might not have heard.

在你常用的播客平台搜索《All Songs Considered》，收听我们精选的2025年最佳音乐。

Search for All Songs Considered wherever you get podcasts to hear us run back some of the best of the best of 2025.

关注时事固然重要，但当你需要放松时，《All Songs Considered》——NPR的音乐播客——随时为你提供陪伴。

Making time for the news is important, but when you need a break, we've got you covered on All Songs Considered, NPR's music podcast.

把它想象成一档音乐发现节目、一场与朋友共度的惬意逃离，当然，还有深刻的音乐洞见。

Think of it like a music discovery show, a well deserved escape with friends, and, yeah, some serious music insight.

我坚持真实。

I'm a keep it real.

我不知道

I have no idea what

这个故事讲的是什么。

the story is about.

每周二，你可以在任何播客平台收听《All Songs Considered》的新集。

Here are new episodes of All Songs Considered every Tuesday wherever you get podcasts.

关注时事，不妨收听NPR的《State of the World》播客。

You care about what's happening in the Stay informed with NPR's State of the World podcast.

只需几分钟，我们就带你了解全球各地的故事。

In just a few minutes, we take you to stories around the globe.

你可能会听到全球冲突的最新进展，或是全球事件如何影响你的咖啡价格。

You might hear the latest developments in world conflicts or about what global events mean for the price of your coffee.

收听

Listen

到

to

来自NPR的《世界现状》播客。

the state of the world podcast from NPR.

这是来自NPR的《就此而言》。

It's consider this from NPR.

2025年对人工智能来说是重要的一年，尤其是人们不断在网上发布的短篇AI生成视频。

2025 has been a big year for artificial intelligence, especially for short AI generated videos that people keep posting online.

孩子们现在称它为AI垃圾。

The kids and now Merriam Webster call it AI slop.

NPR的杰夫·伦菲尔德和香农·邦德今年大部分时间都在这些垃圾中打滚。

And NPR's Jeff Rumfield and Shannon Bond has spent much of the year rolling around in the slop.

他们今天来为大家梳理一下重点，谈谈这背后的意义。

They are here with the highlights, so to speak, and to talk about what it means.

你们好。

Hi to you both.

嘿，斯科特。

Hey, Scott.

哼哼。

Oink.

哼哼。

Oink.

所以我们今天要讨论三个虚假视频，这些视频都在网上广泛传播。

So we're gonna we're gonna talk about three fake video todays, all of which have been widely shared online.

香农，我们先从你开始吧。

Shannon, let's start with you.

我们的第一个AI垃圾内容是什么？

What's our first slop entry?

嗯，去年十月，那是一个星期六。

Well, back in October, this was a Saturday.

那是反对特朗普政府的大规模‘无国王’抗议活动的日子之一，对吧？

It was one of the days of those big no kings protests, right, against the Trump administration.

特朗普总统发布了一段他驾驶战斗机的AI生成视频。

President Trump posted this AI generated video of him flying a fighter jet.

飞机侧面写着‘国王特朗普’。

The jet says king Trump on the side.

他戴着一顶王冠。

He's wearing a crown.

他飞过一座满是抗议者的城市，向他们倾倒看起来像粪便的东西。

He flies over a city full of protesters and dumps what looks like poop all over them.

你可能已经看过这个视频了。

You've probably seen it.

确实看过。

Sure have.

而且

And

这段视频配的是肯尼·罗杰斯的歌曲《危险地带》。

the video is set to the song Danger Zone by Kenny Loggins.

顺便说一句，罗杰斯对他的音乐被用在这段视频中并不高兴。

I should say, as an aside here, Loggins is not happy about his music being used in this video.

他早在十月就要求下架这个视频。

He asked for it to be taken down back in October.

但此后它一直未被下架。

It has not since been taken down.

当然，这个视频显然是假的。

Now this video is, of course, obviously fake.

对吧？

Right?

我的意思是，我不认为有人看这个视频时会真以为特朗普在驾驶战斗机。

Like, I don't think anyone's watching it thinking Trump is really flying a fighter jet.

我们之前也见过总统分享过AI视频和图像。

And we have seen the president share AI videos and imagery before.

他在2024年竞选期间确实做过很多次。

He did actually a lot during the 2024 campaign.

他和他的支持者似乎非常喜欢这类表情包。

He and his supporters seem to really love these kind of memes.

但自从他今年上任以来，这种策略不仅被总统，也被他的整个政府所采纳。

But since he's taken office this year, this kind of strategy, it's really something we've seen not just the president, but his whole administration embrace.

白宫和国土安全部的社交媒体账号经常发布这类由AI制作的梗视频和图片。

The White House and the Department of Homeland Security, their social media accounts post these sort of meme videos and images often made with AI.

我认为，考虑到明年我们将迎来中期选举，我们应该预期在我们的信息流中看到更多由AI生成的政治内容。

And I think, you know, what this tells us, given we're heading into midterm elections next year, we should expect to see even more AI generated political content all over our feeds.

我的意思是，香农，这种情况太普遍了。

I mean, Shannon, this is so prevalent.

你看到特朗普和他的盟友越来越频繁地转发这些内容。

You're seeing Trump and his allies return this more and more and more.

白宫对它们使用AI制作这些视频——我觉得可以说，这几乎就像宣传视频——有什么回应吗？

What has the White House said about their use of you know, I feel like it's fair to say that it's almost like propaganda videos that they're creating with AI.

是的。

Yeah.

我应该说清楚。

And I should be it should be clear.

并不总是清楚这些内容是由白宫本身或白宫工作人员制作的。

It's not always clear that the White House itself or the White House staff are making this.

在某些情况下，比如这则战斗机视频，我们看到总统或政府账号在转发网上其他人制作的内容。

In some cases, like this fighter jet video, we're seeing, you know, the president or administration accounts resharing content that's been made by other people online.

白宫通常不会对这些具体视频发表评论，但过去他们曾就这种社交媒体策略总体表示过，比如‘梗图还会继续’。

You know, the White House doesn't tend to comment on these specific videos, but what they have said in the past in general about this social media approach, You know, they said sort of things like, you know, the memes will continue.

这显然是一种他们认为能与受众产生共鸣的沟通方式。

It's clearly a a form of messaging, I think, they think resonates with their audience.

而且，你知道，这是一个非常在线的政府。

And, you know, look, this is a very online administration.

这是一个非常在线的总统。

This is a very online president.

他们正在积极融入当前网络上的语言风格，而AI正变得越来越重要。

And this is you know, they're very much engaging in the language of what is online at the moment, and it's increasingly becoming AI.

所以，这就是我们正在讨论的第一个趋势。

So that's the first trend we're talking about.

我们来看第二个视频。

Let's move on to video number two.

它来自一家小公司。

It came from a little company.

几乎没人听说过它。

Nobody's really heard about it.

我们之前没怎么讨论过它。

We haven't talked about it that much.

它叫OpenAI。

It's called OpenAI.

去谷歌搜一下。

Google it.

实际上，它展示了这家公司的首席执行官犯下罪行。

So it actually it showed the company's CEO committing crimes.

杰夫，这到底是怎么回事？

Jeff, what's going on here?

是的。

Yeah.

第二个视频来自OpenAI今年秋天推出的一款名为Sora的应用程序，这款应用让生成AI合成内容变得异常简单。

The second video came from an app OpenAI rolled out earlier this fall called Sora, and that app has made AI slops super easy to generate.

Sora的一个功能是，你可以在获得他人许可的情况下，将他们的面容和声音放入你的视频中。

One of the features of Sora is you could put other people's faces and voices into your video with their permission.

最早授权使用自己形象的人之一是OpenAI的首席执行官萨姆·阿尔特曼。

One of the first people to grant permission was Sam Altman, OpenAI's CEO.

他允许人们用他的形象制作视频，一名OpenAI员工制作了这段阿尔特曼似乎身处靶场的视频。

He let people make videos with his likeness, and an OpenAI employee created this video of Altman in what appears to be a target.

这是监控录像，阿尔特曼看起来正在为他的AI公司偷窃计算机芯片。

It's surveillance video, and Altman seems to be shoplifting computer chips for his AI company.

求求你，我真的很需要这个用于Sora推理。

Please, I really need this for Sora inference.

这个视频太棒了。

This video is too good.

这其实是一个关于AI对计算资源无尽需求的内部笑话，但它在研究上确实值得注意。

Now this is an inside joke about AI's endless need for computing, but it's really notable for a couple of research.

首先，它表明AI视频现在可以将真实人物放入完全虚构的情境中。

First, it shows that AI videos can now put real people into completely fake situations.

你可以让一家公司的首席执行官犯下虚假的罪行，而且看起来非常逼真。

You can make the CEO of a company commit fake crimes and make it look pretty real.

但Sora能够生成的虚假内容还不止这些。

But that's not the only fake stuff that Sora is capable of producing.

所以，你知道，我们也看到过一些新闻报道，称Sora生成了人们伪造投票箱、虚假地方新闻采访的视频。

So, you know, we've also seen news stories about Sora producing fake videos of people stuffing ballot boxes, fake local news interviews.

这引发了大量担忧，特别是，正如香农刚刚提到的，明年我们将进入选举年，而Sora基本上已经将生成劣质内容的门槛降到了零。

And this is creating a lot of concerns, especially, you know, as Shannon just said, we're going into an election year next year, and Sora has basically lowered the bar for slop to zero.

所以，杰夫，这里存在两种截然不同的极端情况。

So that's two different extremes here, Jeff.

你谈到的是这些虚假视频如何能迅速进入真实新闻循环。

You're talking about ways that these fake videos can get into the real news cycle very quickly.

香农，你所说的这种东西，简直可以说是荒诞不经的宣传，找不到别的词来形容了。

Shannon, you're talking about just totally farcical propaganda in in it, for a lack of another word.

我们来看最后一个视频。

Let's move on to our final video.

这个视频是其中之一。

This is one.

也许我们的听众已经看到，它在TikTok上获得了大约两亿次播放。

Maybe our listeners seen it's racked up, like, 200,000,000 views on TikTok.

香农，跟我们说说这些兔子吧。

Tell us about the bunnies, Shannon.

是的。

Yeah.

这个视频看起来也非常逼真。

So this video, it looks also pretty realistic.

它看起来像是夜间后院蹦床上几只非常可爱的兔子在跳跃的环形摄像头画面。

It looks like Ring camera footage of some very cute bunnies bouncing around on a backyard trampoline at night.

你可以想象，人们对这些视频的发布方式，就是从他们真实的Ring摄像头拍摄的。

And you can imagine, right, people post these kinds of videos, right, from their actual Ring cameras.

所以今年夏天，当这个视频在TikTok上发布时，很多人被误导，以为它是真实的。

And so when this was posted on TikTok this summer, a lot of people were fooled into thinking it was real.

视频本身没有任何水印，表明它是用AI制作的。

There was no watermark on the video itself, disclosing that it was made with AI.

TikTok后来在该帖子上添加了自己标注的AI标识。

TikTok has since put its own AI label on the post.

但你知道，从当时评论来看，确实有大量人完全相信这是真实的。

But, you know, judging from the comments at the time, you know, lots and lots of people just totally thought it was real.

我觉得这个视频挺有意思的。

And I think it's interesting about this one.

你知道，它和我们之前讨论的其他例子很不一样。

You know, it's quite different than the other examples we've talked about.

但这种视频正是那种无意义的、可爱的、用来吸引互动的 bait。

But this is the kind of, you know, mindless, cute engagement bait.

是动物。

It's animals.

对吧？

Right?

这在互联网上非常普遍。

That's so prevalent on the Internet.

一直以来在互联网上都很普遍。

It's always been prevalent on the Internet.

对吧？

Right?

所以某种程度上，我们现在看到这种AI版本并不令人惊讶。

And so in some ways, it's not surprising that now we're seeing AI versions of this.

但让我感到震惊的是，我现在在社交媒体动态中到处都能看到这种内容。

But what strikes me is this is the kind of stuff I am seeing all over my social media feeds at this point.

无论它们是否被明确标注为AI生成，这确实开始模糊了界限，让人觉得，如果上网，就无法逃避这种AI垃圾。

And whether or not they are, like, clearly labeled as AI, it really does start to blur the boundaries, and it makes people feel, I think, like this AI slop is inescapable if you are going to be online.

如果这是无法避免的，我只是想知道，杰夫，我们能做些什么来应对这种情况吗？

And if it is inescapable, I'm just wondering, Jeff, is there is there anything we could do about that?

我的意思是，首先，在出现某种监管和标识之前，斯科特，你可能只能接受自己迟早会被骗的事实。

Well, I mean, the first thing is, you know, until there's some sort of regulation and labeling, you're probably just gonna have to accept, Scott, that you're gonna be duped sooner or later.

我的意思是，到目前为止，我们所有人都看过一些AI制作的视频。

I mean, I think all of us at this point have seen videos that are AI.

但话虽如此，还是有一些需要注意的地方。

But that being said, there are some things to watch out for.

AI视频通常都很短，嗯。

AI videos tend to be very short Mhmm.

因为制作它们需要大量计算资源，而且它们常常包含一些场景，如果你稍加思考，就会发现这些场景相当不真实。

Because it takes a lot of computing to make them, and they often contain scenarios that if you take a second, you'll realize are kind of unrealistic.

比如，那些钱不可能同时在蹦床上弹跳。

Like, all those monies aren't gonna bounce on a trampoline all at once.

反向图片搜索也有帮助，或者你可以搜索你所看到事件的新闻报道。

A reverse image search can help too or searching a news story on the event you're seeing.

但有趣的是，斯科特，我所接触的一些研究人员表示，他们并不希望人们变得愤世嫉俗，进而认为一切都是假的。

But interestingly, Scott, you know, one of the things researchers I spoke to about this say is they actually don't want people to become cynical and just assume everything is fake.

因为一旦这样，就很难让那些恶意行为者承担责任了。

Because when that happens, it makes it really hard to hold bad actors to account.

你知道吗？

You know?

人们会说，哦，那只是假的。

People can say, oh, that's just fake.

我根本没做过那件事。

I didn't really do that thing.

所以我们必须努力坚持现实，即便是那些可爱的动物视频也不例外。

And so we've gotta try to cling to reality even for the cute animal videos.

你知道吗？那只在马桶旁晕倒的浣熊？

You know, that raccoon that passed out next to the toilet?

我本来以为那是AI生成的。

I thought that was AI.

但我做了功课，感到如释重负，看到2025年一只浣熊真的能在酒铺喝醉，这给我带来了一点快乐，这确实是真实的事。

But I did my homework, and I'm was relieved, and and it brought a little joy into my life to see that a raccoon really can still get drunk in a liquor store in 2025, and that's that's a real thing.

我认为，我们总能指望浣熊为我们提供真实又有趣的网络内容。

We can always count on raccoons to give us realistic entertaining Internet content, I think.

但愿如此。

Let's hope.

但愿这些浣熊不会因为所有的AI图像生成而失业。

Let's hope the the raccoons aren't put out of work by all this AI image generation.

这是NPR的香农·邦德和杰夫·布姆菲尔德。

That was NPR's Shannon Bond and Jeff Brumfield.

深陷泥潭啊。

Deep, deep, deep into the slop.

谢谢你们，也向你们两位表示哀悼。

Thank you, and my condolences to both of you.

谢谢，斯科特。

Thanks, Scott.

谢谢。

Thank you.

本集由埃琳娜·伯内特和丹尼尔·奥夫曼制作，由约翰·凯奇姆、布雷特·尼利和科特妮·多宁编辑。

This episode was produced by Elena Burnett and Daniel Offman and edited by John Ketchum, Brett Neely, and Courtney Dorning.

我们的执行制片人是萨米·叶尼根。

Our executive producer is Sammy Yenigan.

感谢我们的《考虑此》付费听众，你们支持NPR记者的工作，帮助维持公共广播的活力。

Thank you to our Consider This Plus listeners who support the work of NPR journalists and help keep public radio strong.

支持者还能在无广告干扰的情况下收听每一集，并解锁《考虑此》的独家附加集。

Supporters also hear every episode without messages from sponsors and unlock bonus episodes of Consider This.

了解更多请访问 +.npr.org。

You can learn more at +.npr.org.

这是NPR的《考虑此》。

It's consider this from NPR.

我是斯科特·德特拉。

I'm Scott Dettra.

跟上新闻可能感觉像一份24小时不停的工作。

Keeping up with the news can feel like a twenty four hour job.

幸运的是，这正是我们的工作。

Luckily, it is our job.

在NPR新闻速递播客中，每小时我们都会将最新、最重要的新闻整理成五分钟的短篇，让你在会议间隙或赶路途中轻松收听。

Every hour on the NPR News Now podcast, we take the latest, most important stories happening, and we package them into five minute episodes so you can easily squeeze them in between meetings and on your way to that thing.

现在就收听NPR新闻速递播客。

Listen to the NPR News Now podcast now.

NPR的播客《特朗普的任期》是您获取特朗普政府重大新闻当日更新的来源。

NPR's podcast Trump's terms is your source for same day updates on big news about the Trump administration.

每集短小精悍，聚焦一个主题，大约五分钟。

Short, focused episodes, one topic at a time, about five minutes or so.

我们整合NPR各领域的报道，确保您始终获取最重要、最紧急的新闻。

We carry out reporting from across all of NPR's coverage, so you are always getting the biggest, most urgent stories.

请在NPR应用或您常用的播客平台收听《特朗普的任期》。

Listen to Trump's terms on the NPR app or wherever you get your podcasts.

想听没有广告插播的这个播客吗？

Wanna hear this podcast without sponsor breaks?

亚马逊Prime会员可以通过亚马逊音乐免费收听本播客无广告版，或者您也可以通过访问+.npr.org支持NPR的重要新闻报道，获取无广告版的《Consider This》。

Amazon Prime members can listen to consider this sponsor free through Amazon Music, or you can also support NPR's vital journalism and get consider this plus at +.npr.org.

网址是+.npr.org。

That's +.npr.org.