NotebookLM刷屏后，我们总结了10个用法并重新回答一个老问题：AI如何影响内容？

欢迎收听 《 十字路口 》， 我们关注新一代 AI 技术浪潮带来的行业新变化和创业新机会 。 十字路口是乔布斯对苹果公司的一个比喻 ， 形容它站在科技与人文的十字路口 ， 伟大的产品往往诞生在这里 。AI 正在给各行各业带来改变 ， 我们寻找 、 访谈和凝聚 AI 时代的 " 积极行动者 "， 和他们一起探索和拥抱新变化 、 新的可能性 。

我是主播 Koji 杨元成 ， 联合创办了 《 街旁 》《 新世相 》 和 《 躺倒 》。 我相信科技 ， 尤其是 AI， 会在未来 10 年彻底改变社会 ， 赋能人类 。

欢迎大家找我聊天 ， 碰撞想法 ， 链接下一个可能性

。

我是主播 Ronghui， 目前在一家专注科技投资的风险投资机构工作 ， 之前在 《 第一财经周刊 》 担任驻硅谷记者 。

在本周节目正式开始前 ， 照例是我们的 " 十字路口黑板报 " 栏目 。在这个栏目中， 我们将邀请朋友们来介绍他们的最新产品 、 活动 、 招聘等等动态 。

上周我们讲了两个朋友的新项目 ， 这周我们自己也有两件 AI 新鲜事儿跟大家分享 。 第一件是我们最近在 《 新世相 》 发起了一个叫做 "AI 遗嘱 " 的产品 ， 邀请年轻人来写遗嘱 。

首先别被 " 遗嘱 " 这个词吓到了 ， 这不是什么不吉利的事儿 ， 恰恰相反 ， 它是一次可以让人去深度思考生命的机会 。

我们希望通过产品去激发年轻人通过直面死亡来思考生命的意义 。AI 遗嘱怎么写呢 ？ 并不是让 AI 来帮你写 ，而是请 AI 来对你进行采访 。AI 很会引导 ， 它会不断地追问细节 ， 追问你在答案背后的深层思考 。AI 大概会提出 10 个左右的问题 ， 之后 AI 将会根据对话整理出一篇完整的遗嘱 。

目前已经有大约 1 万人在 AI 的帮助下写下了自己的遗嘱 ，也欢迎你来试一试 。 通过写遗嘱 ，在与死亡四目相对之后， 我们或许能够看清人究竟在为什么而活 ， 从对死亡的思索当中获得对生的领悟 。

今天的第二条黑板报 ，是我知道这个十字路口的朋友当中不少都在海外创业 ， 尤其是在硅谷创业 。

大家多多少少都在考虑利用好中国优秀的研发资源 ，在国内设立一个研发团队 。 因为不但工程师的薪资低 ，而且沟通与交付的质量都很好 ，但是如果创始团队在国外， 要远程在国内来招聘管理 ， 甚至哪怕是发工资 ， 交五险一金 ， 都挺繁琐的 。

所以最近我有朋友开始在系统性地提供这项服务 ， 帮助了 20 多家在硅谷的 AI 公司在国内招聘和管理研发团队 。他们在北京 、 上海和深圳都设立了办公室 。

如果你有类似的需求 ， 欢迎在 show notes 当中添加十字路口的小助理的微信来了解更多 。 同时， 如果你是正在寻找新工作的工程师 ，也建议你可以考虑一下远程去为硅谷的 AI 公司工作的机会 ，不但可以赚美金 ，而且如果你加入的是华人创办的公司 ， 我想语言沟通的问题应该也不会太大 。

感兴趣的工程师的朋友们也欢迎在 show notes 里面找到十字路口小助理的微信进行添加 。

大家好 ， 欢迎收听这期的 《 十字路口 》。 这期播客是久违的我们俩聊天的形式 。其实最初是我们俩想聊一聊中秋节前我去香港参加声动活泼主办的 Future&Remix 活动 ， 特别是我在其中主持了一个关于 "AI 如何改变我们所消费的内容 " 的一个圆桌讨论 。

这个圆桌讨论的出发点是 ，AI 显著地降低了内容创作的门槛之后， 让更多的人有机会成为创作者 ， 生产出更多有趣的内容 。

这必然会给信息消费 、 供给和生产带来很多变化 。 很有意思的是 ， 就是在这个活动之后， 我们觉得 AI 和内容这个领域出现了一个新的 、 可能会带来非常多改变的一个非常重要的产品 NotebookLM。

所以我们今天就决定从 NotebookLM 助手来聊一聊我们的使用体验 ， 我们看到了它的一些使用场景 ， 和站在今天来看这个 NotebookLM 出现 ，并且看起来它正在朝着一个杀手级应用的方向飞奔 。

我们现在站在今天这个角度上， 又要怎么来回答这个 panel 的问题 。 首先我们先来请 Koji 来给大家简单介绍一下 NotebookLM， 虽然很多朋友其实可能已经用了 ，但是对还不了解 NotebookLM 的朋友先来快速了解一下它是什么 。

NotebookLM 是 Google Labs 出的一款产品 ， 然后它其实说来和它的名字是一样的 ， 是一个 notebook， 是一个笔记本 。 然后你可以把你在某个主题之下的所有的内容全部丢给它 ， 然后向它提问 ， 请它生成提纲 ， 甚至生成一段音频 。

比如说如果你最近在研究某种小鸟 ， 比如说你在研究夜鹭 ， 那你就可以把夜鹭的 10 个 PDF、100 个音频 、10 个视频全部都丢给它 。

然后这个时候你可以请它用它最牛逼的 ，也是在网上最广为流传的那一个功能 ， 就生成一个音频 。 这个音频是两个播客主播在这儿一本正经 ， 然后又非常有活力地 、 轻松地针对你的内容来聊天 。

然后它之所以能够刷屏 ， 就是因为这个音频它自然 、 准确 、 充满活力 ， 讲得比真人还要好 。 然后我其实觉得这么介绍吧 ，有一点就是难以用语言表达它有多好 ， 所以我们给大家做了一个例子 ， 就我们把十字路口这档播客的介绍 ， 还有我们历史上做过的一些内容 ， 全部丢给了 NotebookLM。

它有一个特别强的地方 ， 就是它支持超强的上下文窗口 ， 所以丢给它的内容再多 ， 它都能处理得还蛮好的 。

然后接下来大家听到的这一段音频 ， 这一段二人播客 ， 就是 NotebookLM 生成的对十字路口的介绍 ， 大家听听看 ， 可以感受一下 。

Ever get the feeling like we're standing on the edge of something massive? Well, today we're diving into that feeling, specifically the world of AI, and we're doing it through this Chinese podcast called Crossing.

I think what's really interesting about Crossing is it really focuses on the intersection of AI with business and entrepreneurship. It's about making AI relatable, accessible, even exciting for those trying to ride the wave of this tech shift.

And it's not just another dry technical podcast. The name itself, Crossing, seems to suggest a bridge between the complexities of AI and this very human desire to grasp it and use it.

Exactly. The show, it cleverly draws this parallel with how Steve Jobs described Apple standing at the crossroads of technology and humanity. And Crossing is suggesting, I think, that it's in that convergence, it's where things get really interesting.

I like that. So it's about finding the human story within the code. So we've got show notes, episode descriptions, all that good stuff. What's the core philosophy, would you say, that's driving this podcast?

So at its heart, I think Crossing is driven by this very strong belief that AI is a force for good. It's a tool that has the power to empower individuals and revolutionize how industries work. And they're very intentional about emphasizing the human element.

Oh, how so?

Well, they actually talk about their human touch. It's how the host's personality shines through, how they really delve into the emotional and societal impacts of AI. They'll explore themes of career shifts, financial uncertainty, even the future of human connection in this AI-driven world.

Wow.

不知道大家听完是什么感受 ， 反正我其实会觉得很多时候比我说的好多了 。

我是觉得他们说的百分之百都比我好 。

那到今天 Koji 你会觉得它是一个 Killer app 吗 ？ 因为其实继 ChatGPT、Character.ai 之后， 好像 NotebookLM 是一个最新出现的一个相对来说更现象级的一个产品 。

但是另外一个点是 ， 我们站在今天这个时间点 ， 现在已经 10 月了 ， 这个产品其实出得挺早 ， 它是 9 月的时候才被更多的人看到了 。

你当时用跟现在用的感受又有什么不一样呢 ？

我觉得首先这个产品是迅速地刷屏的 ， 尤其是它用生成的音频 ， 大家真的都叹为观止 ， 自然就传播起来了 。

然后我自己感觉这一个月过去看到了越来越多的用例 ， 所以今天其实我们也准备了 10 个我们看到的 NotebookLM 大家用得比较特别的 、 又有价值的一些 use case 来和大家分享 。

这是一方面 ， 待会我们会给大家讲这个部分 。 然后另外一方面是 ，其实这个一个月 NotebookLM 还是在不断地升级 ， 比如说它现在支持了团队 ， 支持了公司的协作 ， 这个我觉得也是蛮重要的一个升级的 ，因为大家共同做一个项目这件事情 。

然后这一个月其实有非常多关于 NotebookLM 怎么用的文章和视频在网上到处的流传 ， 然后这里面有一些是大家想得到的 ，有一些是非常搞笑的 。

然后我们今天会和大家分享 10 个 ，有一种抛砖引玉 ， 给大家一些启发 ， 给大家一些看看别人在怎么用的这样的参考 。

比如第一个 ， 我们先说一个搞笑的 ， 就是有人试着把自己的简历丢进去 ， 然后请这两位主播生成一段夸夸播客 ， 然后你可以听到两个人对你的简历当中的各种细枝末节都表现出强大的兴趣和浓厚的情绪价值 。

甚至有人讲说 ， 听这两位主播如此热情洋溢地讨论我 、 表扬我 ， 这个效果简直堪比去看了一次心理医生 。

然后第二个 ， 看到的用例是非常质朴的 ， 就是把会议记录丢给它 。 然后因为大家知道在公司越大 、 组织越大 ，有时候会议就越冗长 ， 然后现在会议也都有录音 ， 然后有时候老板也会要求说 ， 今天你没有来这个会 ， 那你下来把这个录音听一下， 至少你要 catch up。

所以 NotebookLM 就提供了一个可能是 ， 你可以把两个小时的会议音频丢给它 ， 它给你生成一个 5 分钟或 10 分钟的二人对谈的播客 。

这个对谈就听起来肯定比领导发言有趣多了 ，因为大家在刚才也听到了那两个人是表达能力也好 、 幽默感也好 、 这种情绪强度也好 ， 让你听得下去 。

这是第二个用法 。 然后第三个用法是我看到有人把一本 《 红楼梦 》 整个丢进去 ， 然后或者把一本 《 百年孤独 》 整个丢进去 ， 然后请 AI 来讲书 。NotebookLM 生成的讲书的内容也挺好玩的 ， 这个是第三个用例 。

然后第四个是我看到国内的产品 ，Monica 他们出了一款新产品 ， 这是一个 APP， 这个 APP 叫做 Riff Podcast，RFF Podcast。 它其实和 NotebookLM 是相关 ，是被他们启发 。

它做的事情是什么呢 ？ 它会把 Product Hunt 或者 GitHub Trending 或者 Hugging Face 或者 Arctify 上面每天的热门的内容抓取下来 ， 然后再生成一段播客 。

这样你就像每天听晨间新闻一样 ， 只是你听的晨间新闻是当天由 AI 去抓取的你感兴趣的信息源的当日的热门内容 ， 然后生成了一段听起来像模像样的 、 品质蛮高的一个播客 。

我觉得这个是一个

非常迅速的 、 有执行力的一个落地 。其实很多朋友听到 NotebookLM 之后有一个反应就是 ， 那我是不是可以让它取代以后每天早上的国际新闻或者国内新闻这样的早间广播 ？

理论上是的 ， 只是今天还没有人去认真地做 。 我看到 Monica 团队他们做了这么一个产品 ， 还是蛮感慨的 ，但他们选的场景是比较垂直的 ，是针对创投行业或者做 AI 领域的朋友们感兴趣的内容去做的产品 。

这是一个超越了 NotebookLM， 就它启发出来的新产品 。 这是第四点 。 我看到第五个有趣的应用的场景是 ，因为大家知道 AI 大神 Ilya 推荐了非常多的论文 ，而看论文这个事情其实还是蛮枯燥的 ， 所以就有 Ilya 的粉丝把 Ilya 曾经推荐过的所有的论文一起丢给了 NotebookLM， 然后请它生成了 Podcast，并且还把这个 Podcast 上架到了 Spotify， 现在一共有 30 期 。

这个 Podcast 的名字叫做 《The AI Research》， 大家感兴趣可以去听一下 。 我觉得这其实也是某种意义的科普 ， 就直接把论文变成科普的小音频 ， 让大家听起来更加的轻松 。

Andrew Caparti 他自己也做了一个类似的 ，但不过这个应该是他自己做 ，因为他也是 NotebookLM 的超级支持者吧 。他是把一些很深奥的那些书 ， 然后用这些书作为原文档 ， 然后让 NotebookLM 生成了一个播客 。

那个播客挺深刻的 ，也在 Spotify 上 。

对 ，其实它这背后我觉得包括 《 红楼梦 》《 百年孤独 》 还有这些论文 ， 它们都类似 ， 就是把一些大家读起来很费劲 ， 甚至有点读不下去的一些内容 ，但大家又感兴趣 ， 又觉得有价值的内容 ， 把它换一种形式来去解读 、 来去表达 。

过去其实也有人讲书 ， 对吧 ，也有人去讲论文 ，但是今天 NotebookLM 出现之后， 你可以就接近零成本的 ， 至少今天它也没有收费 ， 接近零成本的把这些枯燥的 、 冗长的 、 你觉得该看但没时间看的内容 ， 把它变成一个轻松的可以消费的音频 。

所以这是 NotebookLM 让很多人感到兴奋的一个主要原因 。 然后我们要分享的第六个案例 ， 它有点像 ， 它是有一个人把一家银行在过去三年的财务表现 ， 你可以想象这个有 ， 如果你要去看那个 PDF 的话 ， 得有几百页一年， 对吧 ， 这三年的长达数千页的 PDF 全部丢给 NotebookLM， 然后请它来去讨论这个银行在过去三年财务表现到底怎么样 。

所以这也是一个大开脑洞 ，但是听一听又觉得能非常强大地提高我们的生产力 ，以及让我们可以一个人去获取更多的有价值的 、 有洞察的信息的一个好做法 。

荣蕙你看到了些什么 ， 可不可以也分享一下 ？

对 ， 我看到一个我觉得一个非常直接跟可能大家非常熟悉的一个应用场景 ，Richard Martin 自己在推特上分享的 ，他最近自己用的一个方式 ， 就是他去圣天使城堡旅游 ， 然后他找了很多关于圣天使城堡的一些资料 ， 上传到 NotebookLM， 让 NotebookLM 给他生成了一个 ， 就是他们那个音频 deep dive， 然后他自己一边游览 ， 然后他一边听 NotebookLM 给他生成了旅游景点的介绍 。其实这个我觉得它跟现实场景有

非常相关的一个映射 ， 就很像我们在比如说在博物馆里面听的导览 ， 然后他说他自己就沿着路边就一边走一边听自己的音频 。

之后他还写了很多就是他自己的一些反馈 ， 就比如说他听完之后觉得说一定要还是需要一个移动版的会更方便等等等等。

如果大家有兴趣的话 ，其实可以去看一下他当时写的这个 。 这是等于是我们列的第七个应用 。 然后第八个是有一个公司叫做 Sardine， 它是用公司的 200 页的文档给 NotebookLM， 然后让 NotebookLM 生成了一个 10 来分钟的一个对话 。

这个公司应该是一个做安全的公司 ， 然后他们这个对话是讲这个产品是怎么样来保护用户的 ， 保护他们的客户的银行账户的安全 。其实这个用例是一个非常典型的 product marketing 的一个应用 。

比如说我们其实也看到很多公司会做播客来讲自己的产品 ， 来讲自己的应用 ，其实这就是一个可以说是成本非常低的一个来做 product marketing 的方式了 。

第九个是 NotebookLM 团队中有一个就是让特别的人， 这个等会我们也会讲到 ，以前是一个记者 ， 是一个作家 ， 叫 Steven Johnson。

我看到他在接受一个媒体采访的时候说他自己是怎么用 NotebookLM 的 。他这个故事好像已经在好几个采访里面都讲到 ， 就是他会用 Readwise 做读书笔记 ， 然后他一年差不多能写 8000 条 。他自己有一个叫做 Nextbook 的 notebook， 然后他会在里面放一些他对下一个项目 、 下一本书的一些计划 、 一些灵感 ， 然后素材 ， 然后让 Notebook 来帮他头脑风暴 ， 比如说让 NotebookLM 给他建议这个文章的结构应

该怎么样来设计 ， 然后根据他上传的材料来提供一些洞察 。他自己就说这个东西很像他的一个 personalized 的助手 。其实这个就让我想到其实以前采访老罗的时候 ， 老罗就说为什么他以前演讲的时候会有那么多的东西讲 。他就说其实他自己有一个笔记本 ， 这个手机上的一个笔记本 ， 然后他会想到什么 ，他就往这个笔记本里存一些东西 。

所以当他需要用的时候 ， 等于就是他就有一个素材库 。 就我感觉这是一个古早互联网时代的一个 notebook 吧 。

第十个用例就是我们找了一个我觉得还是挺开脑洞的这么一个想法吧 ，是在 Reddit 上看到的 。 然后在 Reddit 上有一个人说 ， 就是他用 NotebookLM 来重新看自己以前看的书 ，他自己以前写的博客 。他说他的感受就是 NotebookLM 把他的从前的那些经历 ， 从前的那些阅读给复活了 。

因为如果不是 NotebookLM 的话 ，他可能都已经把这些都给忘了 。 所以他就很好奇 ， 会不会有一种可能就是当我们把以前的一些资料全部都丢给 NotebookLM， 然后让它来做成播客每天播出 ， 这样就会变成一个人的人生的楚门的世界 ， 等于就是每天都在重新地看自己的生活 。

然后让这个分享更有意思一点的是下面的第一条评论 ，有一个人说 ， 你可以再配上一个 Meta Ray-Ban 的眼镜 ， 然后每天拍你周围的东西 ， 每隔 15 分钟拍一些照片 ， 然后用一些模式来记下你每天吃了什么 ， 这样你就是在实时地存档你每天在干什么 。

那就得慢慢地生成自己的人生传记 。 以上就是我们和大家分享的看到的好玩的 10 个 NotebookLM 的应用 。 当然绝大多数的应用我们没有讲到的 ，其实是更加正统的 ， 就大家用它来学习 、 做科研或者用它来工作 。

我们分享的其实是一些更有趣 、 更好玩的一些开脑洞的用法 ， 还是比较建议大家自己去试一试 。 目前是免费的 。

荣蕙其实我们看到关于 NotebookLM 的评论报道是铺天盖地的 ，在过去两周 。 但其实十字路口我们之所以决定在此刻做这个内容 ，也是想说等这个热浪稍微过去一点 ， 我们可能可以做得更有价值以及更有思考 。

一方面是我们可以看到更多大家在怎么用 NotebookLM 的案例 ， 另一方面是我们可以拉开一点时间线之后， 再去过了那个热潮之后， 再去相对全面地冷静地看这个产品它到底提供了什么价值 ，以及这个产品它到底是在怎样一个环境里面 ，在怎样的一个团队下面 ， 用怎样的工作流程给生产出来的 ， 就做出了如此一个现象级的产品 。

所以在这些所有的评论报道 ， 我们看了那么那么多下来 ， 荣蕙你自己印象最深刻的是什么呀 ？

当时看 NotebookLM 的报道 ，其实一个最直观的感受就是 ， 首先会条件反射地以为这是一个创业公司的项目 ，但是看到它其实是 Google Labs 的其中的一个测试的一个产品 。

我其实也看到挺多人开玩笑 ， 就是说因为这个产品 ， 然后很多人就说可以看到 Google 这个创新的能力还是很强什么的 。

然后我其实去看了一下 Google Labs 的一些介绍 ， 就是它其实是 Google 内部的一个孵化器 ， 用来测试和公开演示一些新的项目 。

它其实在 2002 年到 2011 年是一直存在的 ， 然后到 2011 年后被关闭了一段时间 。 然后在 2021 年的时候 ， 这个名字变成了 Google 内部测试 AR/VR 的一个内部的一个小组 。

然后再到 2023 年，Google 在 Google I/O 大会上又重新宣布他们又有 Google Labs 这个组织 ， 它的工作是 Experiment with the Future of AI， 意思就是说来做未来 AI 的一些尝试 。

如果大家现在去 Google Labs 这个网站上 ，其实看到它有很多个项目 ， 然后 NotebookLM 是其中的一个 。

所以感觉 Google 的创新能力还是非常强的 ， 包括 NotebookLM 还是 Google 一直以来的 80%、20% 的文化之下诞生的 。 它是一个 20% 的时间由 Google 的团队去做创新所做出来的一个产品 。

这个就不得不感慨江海是老的辣 ， 就这样的企业文化和这样的工作方法 ， 它确实是酝酿创新的很好的土壤 。

对 ， 然后在看它的整个组织的时候 ， 发现其实他们应该人还不是特别多 。 看到比如说 Rizal Martin 他上了两个播客 ， 一个是 Lenny's， 然后一个是洪珊的播客 ， 里面都有提到最开始其实只有好像三个人， 然后我后来在 LinkedIn 上看了一下 ，他们应该好像也是十来个人吧 。

对 ， 就做 NotebookLM 的创始团队只有三个人， 到现在就确实人数是非常少的 。 但其实我觉得现在做 AI 产品确实也不需要那么多人了 ， 就这里头有很多基础设施都搭得非常好 ， 然后同时大家有了写代码的 AI 工具 ， 做设计的 AI 工具 ， 一切都超级个体了 ， 都可以变得更高效 ， 团队可以更 lean。

我自己看到 Google Labs 的这些资料的时候 ，其实会很好奇一个问题是 ，因为在一个 Google 这么大的一个公司里面 ， 怎么样去做一个小的组织 ， 它是按照创业公司的思路来做的 。其实肯定它在组织关系的平衡上是要比一个初创公司遇到的挑战是要更多的 。

里面有一个挺有意思的细节是 ，Google Labs 它现在的负责人 ，他的职位是 Google Labs 的 VP， 然后我在 LinkedIn 上看了一下他的经历 ， 就是他其实在 Google 待了超过 15 年 。他的介绍里面是说他在 Google 做过非常多的产品 ， 比如说 Google Pay、Google Wallet， 可以想象就是他其实是可以帮小的团队来平衡很多公司内部的关系资源的这么一个角色 。

这其实也很不容易 ， 一个在大公司工作了 15 年的人， 还有意愿以及有活力和动力去驱动一个创新产品 ， 多数就是在国内大厂工作了 15 年， 就那个人生状态可能再去做一个 Google Labs 这样的项目有点困难 。

我觉得还是挺感慨的 。其实我看到他们团队里面有另外一个我大为震惊的一个团队成员 ， 是一位作家叫做 Steven Johnson， 然后甚至在很多报道里面都有提到 ，在 NotebookLM 这个产品的诞生过程中， 这位作家 Steven 他是发挥了非常非常大的作用的 。

首先 NotebookLM 它其实还是给知识工作者或者是给创作者提供的一个工具 。在生产这个工具的过程当中， 团队里面如果有一位资深的创作者 ， 那他本身一方面就是可以告诉这个团队我是怎么做的 ， 把我的工作的过程 、 工作的流程和我观察到的同行们的工作过程 ， 把它更好地理解并且翻译成产品 。

这个是一个很特别的配合模式 ， 就直接有用户 ，而且是可能全世界最牛逼的用户之一在你的团队里面 。

然后另外一个就是 ，因为 NotebookLM 它最后输出给用户的也是内容 ， 大家都很感慨说为什么做的音频那么好 。

我觉得这里面应该也有 Steven 的功劳 ， 就他会从一个内容怎么叫好 、 怎么叫美 、 怎么叫让大家觉得有用 、 有趣 、 听完之后有共鸣等等角度给反馈 ， 来去配合产品经理 、 配合工程师不断地调整 。

我其实觉得这个很可能是未来要去做 AI GC， 就生成内容的产品的团队都可以考虑的一种工作模式 ， 就是团队里面除了产品经理 、 除了设计师 、 除了工程师 ， 可以考虑以某种方式引入一位在你们所做的这个领域里面最资深的创作者 ， 让他从 day one 就开始深度地参与整个产品的打磨的过程 。

所以他不但可以保证最后产品做出来是大家目标用户用起来顺手的 ，也可以保证生产出来的产品它在审美上 、 在调性上 、 在质量上可能也是过关的 。

这个是我觉得挺有趣的一个团队搭建的未来的趋势 。

对 ， 就其实我在 Lenny's 跟洪珊的播客里面其实都听到主持人有问 Steven Johnson 他发挥重要作用的地方 ， 然后他的产品负责人其实都提到了有一个 content studio 这么的一个特色 ，但是其实在两个播客里这一点上都好像没有特别多的更详细地说 。

所以我理解这个其实是他们的一个壁垒所在吧 。 回到我们说到前面说在香港的 AI 如何改变我们消费的内容的这个主题 ， 我自己感觉是在 NotebookLM 的影响下重新来看当时讨论的问题的时候 ，其实可以想到当时 panel 中 Yolanda， 她是香港大学新闻传播系的老师 ， 她就提到比如说美国有小型的新闻机构 ，他们现在是做调查性新闻报道的时候 ， 用生成式 AI 来总结美国各州

的政策文件 ， 来让读者来更容易地理解 。 如果让记者跟编辑来完成这个工作 ， 一般会花费的时间跟精力会非常非常地多 。其实我觉得可能一个阶段是用 AI 来总结 ， 另外一个阶段是有了 NotebookLM 之后， 把这些资料全部给它 ， 然后让它来生成 。

这就和我们前面说到的一些 ， 比如说把书或者论文丢给 NotebookLM 让它来生成的用例其实有点像 ， 都是在利用它基于原文档和结构化内容整理的能力 。

这个 panel 里面其实也讨论了 AI 带来的假新闻的问题怎么解决 。 当时圆桌的另外一个嘉宾林亦 ，他是 B 站的 up 主 ， 之前是学机器学习 ， 现在在做和 AI 和科技相关的内容 ， 当然他也用了很多 AI 的工具 。

林亦就有说到我们可以用魔法来打败魔法 ，其中我想一方面是 Yolanda 介绍的可以用 AI 来先做一遍 fact check， 就是事实核对 ， 然后来交给人类接管 。

另一方面就是在 NotebookLM 出现之后， 也许可能会有一个改变 ， 就是它可能会改变我们来了解或者是阅读传统的新闻的一个方式 。

因为大家知道其实传统的新闻写作它有比较严格的要求 ， 对很多人来说其实是会觉得有点无聊的 。NotebookLM 生成的内容第一它基于原文档 ， 第二是它可以把内容整理和转化成大家更容易接受的对话式的方式 。

从源头上它给了用户更多的控制权 ， 就是你可以选你要看哪些内容 。其次是它把这种接受的方式 、 互动的方式也变得更容易接受了 。

也就是我就想到前面 Koji 说的 ，是不是有可能以后我们每天听到的国际新闻都有可能是由 AI 来说的 ，也是一种可能吧 。

当然这个里面它还是相对来说加了一个步骤 ， 就是需要人来进行一次操作 ，但我想说其实这可能是一个所谓魔法打败魔法的一个尝试的方向吧 。

在这里我也想多介绍一下这个圆桌 ， 除了 Yolanda 跟林亦 ， 还有一位嘉宾 Gary，他是美图的 CFO，他也和我们分享了一些美图作为一个生产力工具在 AI 这个时代的尝试 。

这里也想和大家说明一下， 接下来的这个圆桌是英文的 ，因为那天现场主要的交流语言是英文 。 我们也在此特别地感谢声动活泼邀请我们去参加这个活动 。

接下来就是这次活动的原声 。 如果大家有任何关于 NotebookLM 或者是这次活动的主题 AI 如何改变我们消费的内容 ，有想探讨的或者是有问题都欢迎给我们留言 。

我们就开始放原声了 。

Welcome our panelists onto the stage. Really happy to be here, to be the moderator of this panel, AI and the future of media consumption. If you look into their background, this panel is kind of like an industry chant.

Content creator, productivity to provider, observer researcher, and policy researcher.

Yeah, if you count me as a content creator as well. Introduce yourself and tell us about the most noteworthy changes AI has brought to your industry and your job in the past 10 years, especially in the past year.

Start with Yolanda.

Hi everyone, this is Yolanda Ma. I'm a teaching at the General Sam School in University of Hong Kong, and so Vepstern University has been really leading, but we're not that bad either.

I used to be a journalism practitioner. I worked in Reuters and S&P here in Hong Kong. I recently just joined HKU, but I had a very long detour in the past decade. I worked for the United Nations on digital policy.

That's why I was introduced as a policy researcher as well. So in terms of the changes in the industry, I'll be focusing more on the news or journalism industry for the past decade. I would say the industry has been playing a very catching up mode.

And I think the last year, well, let me put it that way. Ten years ago, I was still at Reuters, the news agency. And at that time, we just started looking into AI and had a small, very small R&D team somewhere in North America.

But then at that time, that's like really cool and mysterious, something that only a very few people were looking into. And now I think the journalism industry has been pushed into, okay, we have to do something now. Now that everyone become a content creator, everyone become an information provider, now what do we do?

The old power dynamics of publishers own the information providing service is breaking down. So I think the industry is really struggling with that. Let me stop it here and pass on to the next speaker.

Thank you, Yolanda. My name is Gary and I am the Chief Financial Officer at Meitu. Meitu is obviously a mobile app company. Now we have about 260 million monthly active users globally. One third of each is actually outside of mainland China.

We started pretty much as an app company that tried to become a very simplified version of Photoshop, focusing on helping user to create, edit their face, making it look more fair and remove blemishes and all that. But we first started as a PC application and then in 2011 we become a mobile company.

And in 2013 we actually started our own AI lab called MT Lab. And at that point in time, I think computer vision is the most important AI technology that we have. And we used to sort of track the points on the face so that you can edit the face very easily with just one click.

Fast forward to the past one to two years, I think generative AI has really broadened our company scope into helping creators to do a lot of different applications. For example,right now we have an app called XDesign, which is pretty much like Canva, but it's focusing on the e-commerce side.

So users can very easily create a background for an e-commerce photo. And then we also have another app called Kaipai, which is helping podcasters or video makers to easily edit their scripts. And then the video will actually stitch together.

You can also add some effect to make it look very much like reality show effects, highlighting the points that you want to say, so on and so forth. So I think generative AI has really broadened our business scope into helping the creative economy.

A little bit of myself. I feel very compelled to talk about something related to music because of the other panelists. So in 2020 to 2023, I didn't train a lot of AI, but I trained myself playing classical piano.

And it was actually a very painful process, but ended up I am able to finish the LTCL program. So I was actually quite happy. And I really echo with a lot of the previous panelists' view that learning instrument is still very important.

And there's some human elements that really cannot be replaced by AI at the moment, but in the future, who knows,right? Let me just pass on to Mr. Lam.

Hi guys, my name is Yi and I'm a content creator. So we are running a Bilibili channel and a YouTube channel. My co-founder is on the third row here. And yeah, you can wave too. Yeah. Our Bilibili channel has like one million subscribers, but it's in mainland China.

So that's basically the population of just one street, so it's no big deal. So our YouTube channel has 200,000 subscribers. Our content is basically focused on our software projects. It's just like a present the process of developing it.

And so we tell our development story through our videos. As our channel grows, we have more opportunities to maybe talk with industry leaders and listen to their opinions, maybe make some interviews. So we are also doing some of the jobs of journalists.

So that's basically what we are doing. As for the questions, so 10 years might be too much for me because like 10 years ago, I'm still an undergraduate. So if we talk about the influence of AI on the scope of 10 years on me, it shapes my decision of my major selection because I was still studying at that time.

That's when AlphaGo was so popular and I decided to choose machine learning as my major in my graduate study. For the past year, I think what I'm observing is that content creating process is more and more it's getting civilized.

It's getting more and more like common people are using this tool to present their ideas. Previously, if we want to make some films, we want to make some cool scenes, it requires a huge budget. But now everybody could have their innovation ideas and present it through AI tools.

It might not be like that's high quality, but it's like a change from zero to one. It's like they begin to have this tool to express themselves. What I see is that a lot of interesting contents are pouring into the community.

It might be more challenging for us to select, but it also gives more creative ideas the chance to pop out. So yeah, that's what I'm seeing.

Yeah, I have a personal story to share that I think the biggest change AI has made to me in my life was starting my podcast. I told my colleagues about this joke too. I started a podcast to study AI because I'm afraid of losing my job.

I was a reporter too, so I can echo a lot about what Yolanda said. And so the next part we'll talk about to what extent has AI penetrated the production side of information. I think all of us have a lot of things to say.

So the first question is, please share one to two stories, innovations that AI has made our lives possible that was previously impossible. And also what previously unseen needs have been fulfilled. Yolanda first.

Yeah, I can start. I just kind of said the media industry has been playing catch up game, but just to give credit to some of the pioneers in the industry,right? So there have been some, I wouldn't say revolutionary innovations, but there has been some good attempts.

I'll give two examples. One is more on the news gathering side. And so Reuters had a product, I think in 2017 or 2018, which is called NewsTracer. And it's basically doing something that a normal human being, a reporter, will not be able to do.

Basically scrape 700 million daily tweets from back then it's still a Twitter. And basically it's identifying what are the breaking news. And then so in the news business, the old business model is that the faster, the better,right? So basically use AI, it's able to actually scrape all that data and identify the breaking news and give the leads to the reporters for them to identify things happening like a wildfire kind of happening or things that if it's just a reporter saying the newsroom may not be able to identify very quickly.

So it gives Reuters for an agency that win the competition by time, like something around five minutes to 20 minutes lead time compared to its competitors. So that's not necessarily felt by the audience,right? Because for you, ultimately you get the news.

You may not really care if it's a five minutes difference or a 20 minute difference, but it does matter for the business. And the other example I want to give is from the more production side. And again, that may not be something that the readers actually understand how much has been going on the back end.

It's actually happening a lot in terms of how AI is used to process thousands, if not millions of documents for some news organizations. So there was one

US-based organization. It's kind of a smaller one, but mostly working on investigative journalism. And they were able to actually use GenFAI to summarize the policy documents from all the states across the United States and make them more easily readable by the readers.

So that's another attempt. Basically, that's something if you use a reporter or editor to do, that takes a lot of human time and energy. So one quote, I don't remember which editor was that, but one of the biggest news agencies in the US, he said, use AI as like one of your best interns or like hundreds of interns,right?

They're doing a lot of work for you, but you need to train them properly. So I think these are the two examples. One is on the news gathering side and the other is on the production side, which from the receiving end, you may not really feel that difference, but that's happening for some of the newsrooms.

And I'll touch on the distribution side maybe later, but I think that's actually one of the changes that's happening most. So Ling Yi said now all the normal people can start creating content. So the old business model of journalism is breaking down,right?

We're not only competing with Reuters competing in Bloomberg, but we're competing with all the thousands, millions, billions people out there. So then how do normal people like you and me actually receive the information, access the information? How does the information flow work these days?

That's actually more a question for everyone to figure out.

Okay, so the next few minutes is going to sound a little bit like an affairment because what we do is photo and video editing. And obviously AI has a lot of things that changes the way how we edit photos.

So the first thing, for example, is if there's like a random person you want to remove from the photo, you can use AI just highlight that and then that person will be gone and then the background will be generated as if the photo is

original. And so there are other niche sort of functionalities that most people wouldn't use it, but if you really need it, it would be awesome. For example, like my teeth, let's say it's not very

how do you say in English? It's not very aligned. And initially there's no way for you to fix it unless you wear braces,right? But now our app can actually recreate your teeth and then make it look very nice.

Hairline, you can't really do much about it unless you go to see a surgeon, but again, our app can do it. And for video as well,

we would have thought that initially video could be more representative of how you look like, but now we can add a lot of cosmetics, you can change how you look, and then you just need to edit one photo and then the effect will apply to the entire video.

And no matter how you move, that edit won't go away. It will just stick to your face and then you look perfect for the entire video. So a lot of these things were made possible by AI, but I want to emphasize it's not just generative AI.

Traditional AI is also very important. If you look at the percentage of users that uses one or more of the AI functionalities, it's over 87%. So our app, even though people don't really see it as an AI app, but a lot of users are actually using AI functionalities.

Just like Yolanda said, we are having this opportunity of having more quick processing of contents. I hate my interns, but AI could be resolved more quickly. But I hate AI too because it's so hard to talk with them.

I love the technology, but I hate the way to talk to them. It's a new opportunity. We have this power of having so much tools to have ads our use and we can have quick response from them. Like Gary said, we are having this new technology like image processing and video processing.

My point is we are seeing more and more people are having the capability to present their idea to the world with this new AI tool. What I was saying is from the content creation perspective, but I'm also a software engineer.

So from a software developing perspective, I'm also experiencing this new paradigm shift. Previously, I was focused on computer vision. And when we were developing some new apps or maybe functions, we were thinking, where should we get the training data?

That's actually like a real case. It just happened like two months ago. I was thinking having this app to maybe give you a score like on how you look today. From old perspective, I was thinking, where should we get the training data?

Where should we get those daily look photos and get the scores? But all of a sudden we were thinking, oh no, we can just directly use maybe GPT-4.0 because we are having this common technology base so that we could use it to do what we want.

So that's previously is not possible. But now we are having this common technology base that we can use. We have at our disposal.

I think with the help of AI, the things that previously impossible, demand and supply relationship would be changed. I'll take me as an example. I used to work with one or two designer freelancers out of our organization. Andright now, I don't need to work with them because with the help of Canva and Midjourney, I can do that myself.

So the next question will be, are there any changes you guys seeing in the production relationship and content creation?

Maybe I will start first. Interesting when you mentioned that you used to talk to two designers and you now don't need to talk to them. Maybe they're actually getting 10 more clients because of AI.

Yes.

So this is how we see it because I think AI tools, especially productivity tools, actually enable user to do things in a much faster way. So as any technology advancement,right, there will be some jobs that it's being made redundant, but at the same time, you release a lot of productivity.

And that actually, as a whole, of the labor force, actually increase the productivity. I think especially in our industry, we're seeing that very, very prominently because on one hand, we serve over 4 million e-commerce sellers using our app to create content, create posters, create collaterals to sell their stuff.

But on the other hand, we actually have Zetco, which is effectively the hands in China. And on Zetco, there's over 17 million designers and artists on it. And their activity is also booming. So we're actually not seeing anybody suffered as a whole.

So I think as a productivity app company or AI application company, our goal is to really just try to create better tools so users can be more efficient. And at the end, I think the beneficiary will be consumers because you'll be able to get more details about the things that you want to buy.

You'll get better advertising collaterals that at one point might be personalized to you. So I think there will be productivity growth along the way.

Yeah, maybe I can add to that. From the more news or information perspective, I would say there are two kind of phenomenons happening. One is like if any of you, all of you think about how you access information of how you access news these days,right?

It's very fragmented. We all get our information from different apps, different social media platforms, from your friends on the IMs. So it's no longer like 10 years ago or 20 years ago where, okay, I need this piece of news, I go to New York Times, I need this piece of news, I go to whatever media I've been following,right?

That model has been falling apart. So that actually combined with what we were talking about. So Ling Yi, you used the word of civilized. I would say that's democratized of the content production side of things. So then now that comes with the information overload for all of us,right?

I'm sure there must be a moment for all of us in the past week where you were just like overwhelmed by the information. You don't necessarily know where to get that one piece of information that you really need.

So all these phenomens adding up, that is actually there are two challenges here. One is the challenge for individual users. How do we actually develop our own system to actually get to the information source or get to the piece of information that we actually need,right?

And the other challenge for the traditional kind of publishers or news providers is that how do I meet this new demand from the audience? And I think there is a bit of a dynamics happening. And one example is that now that with generative AI, video content become more popular than before compared to two years ago.

So actually there has been a recent study saying because news content assisted by generative AI has been becoming so popular, the old fashioned media companies and publishers are starting to produce more videos to meet this new demand. So I think there will be kind of more mutually dynamic information ecosystem reshuffle happening in the next couple of years.

And we don't necessarily know where that's going, but changes are definitely happening.

Yeah, I think the formal analysis is great. My observation is like there are more and more contents pouring out and definitely there will be bad contents, not only the good qualities. So that might be a problem. But what I'm trying to deal with this scenario is I'm a lazy guy.

So I will just let it go and I will pick the popular videos. And I believe the selection of the audience,right? The audience will pick the good ones for me, but.

Algorithm picks them.

Yeah, yeah, yeah. Algorithm and the audience. You feed the audience like meat to the algorithm,right? Basically, that's still the choice of the audience.

So we talked about the innovation. Now it comes to the problem. What kind of new problems that AI created for us? I think news industry has been disrupted in the most significant way by the technology in the past 10 to 20 years, especially given the development of AI generated content.

And there are some news about some new media outlets that's using AI to generate their news and founded by users and readers. So I guess Yolanda have a lot of observations about this. So Yolanda, can you give us some examples that you saw the new problems that's created by AI?

Yeah, definitely. So I think ultimately the news industry is a fact based industry,right? We need to find the truth. We need to develop the truth picture based on facts. While AI is not necessarily fact based. If we know AI just enough, we know that that's actually based on statistics,right?

So then it's generating the most probable whatever next word, next picture that based on the data it has, which the data itself is biased in many cases. So then in that sense, that's actually, I mean, creators like the first panel, that's great.

I wish we can embrace AI like they do, but we cannot. There are lots of journalism principles we have to stick to, which to some extent is not something that AI is able to do at the moment,right? So one problem or challenges that's actually come with this kind of a conflict is the overload of misinformation online.

And then this year is a big election year. There is about, I think, I might get the number wrong, but.

40 different countries.

Yeah, thank you. 40 different countries coming up with elections this year. So billions of people are going through elections, but they're also facing the challenges of misinformation around elections and regarding campaigns, regarding different candidates. I think there was a example of, that's a real example of a state, I think it's a wavering state in the US starting getting calls from Joe Biden and it's an audio call.

So basically now it's so easy to make up a 10 minute audio based on 10 seconds of someone's voice,right? We have an AI class, AI course in our school and the students can produce that within one class. We can teach them how to basically produce misinformation.

But knowing that, they actually got a call from Joe Biden and telling them to either vote for him or not to vote for Trump, which end up it's an AI generated audio message and not really from Biden, of course.

So this is coming not just a challenge for journalism industry, but also for broader society. So that comes different challenges and different solutions. We need to think of what about the media or information literacy of the audience. I think all of us are pretty well educated.

We probably know how to identify the source, but what about people who are less educated, who may not have the privilege to actually identify or think about people like our parents. My mom sent misinformation all the time. So how to actually do that kind of literacy training, that's something that needs more ecosystem approach to address that.

Have this problem sparked any discussion in the industry?

Yeah.

Solutions?

Yeah. I think as mentioned in my remarks earlier, it's not just something that journalism industry can address. Actually, the reporters also, I think there are two gaps there. One is reporting or enough discussion about AI accountability. So you need to understand the problem enough to actually cover that and to actually tell the audience about it,right?

So that's one discussion and practice start happening. And the other, maybe touching a little bit on my UN part of work, actually that's one of the big projects we've been working on, how to address election related misinformation and how to have that ecosystem approach.

Basically, we developed a product where we can actually use AI to do the fact checking process. It's assisting. It's not just like only using AI to do the fact checking, but you can use AI to help you do some parts of the fact checking and to identify what are the more likely misinformation and then you have the humans following up.

And then by that, we worked in developing countries and have the local media organizations who have the local networks and the local languages to actually do the further step of human kind of vetting and human fact checking. So then you have the ecosystem of media organizations, AI slash technology and government stepping in and then the general public.

So then at the end, hopefully they can address the problem we are facing all together.

Any input?

Yeah. There is this thing of using magic to beat magic. So the problem is caused by artificial intelligence, then maybe we can use artificial intelligence to help solve this problem,right? We can have this AI agent. We were actually thinking about this.

We can have this AI agent to help classify misinformation and facts. Since it's based on AI, so it can be really cheap and everybody could have it.

I can remember Mark Zuckerberg said what exactly you said before. Gary, do you have any input about this?

Personally, I think this is actually not a new problem. I mean, depending on your perspective,right? Fox News versus CNN, one of each is, depending on your stance, is misinformation. So I echo Yolanda's point is, well, education plays a very big part of it.

And to a certain extent, it's easier to discern what is AI generated versus human touch, sort of misinformation wrapped in the principle of under journalism at this point. But obviously in the future, maybe it's more difficult. But I think things like this will happen anyway.

I mean, look back 200 years ago, there's only maybe one newspaper and then newspaper means it's the authority. Now you have Fox versus CNN. So it evolves over time, but I don't think this is not solvable. I mean, it's a good chance for a new human as a whole to raise your critical thinking capability and try to discern and with the help of magic to fight magic, I think this is also a way to deal with it.

So I'm actually still pretty optimistic about it.

Yeah, I think it's not a war between human and artificial intelligence. It's always a war between human and human. Like human with tools to fight against each other.

Yeah, this is a challenge I think newsroom always facing. I remember like 10 or 15 years ago when the first time I saw the H5 New York Times created. My instinct was this is a big challenge for newsrooms because for newsrooms, they need more and more engineers to help them to create those technological works to attract more users.

But the question is, why do top engineers go to work at New York Times instead of Facebook or Google?

Because they're too much stuff. Sorry. Allright, maybe it's not meant to.

No, it was a joke because there are too many of them. So they will overflow to every industry.

I actually have an additional question. So Gary, have you been reached out by any content creators or media groups to work with?

I mean, in what capacity?

I mean, maybe just interested in talking with you guys to see the potential of working with Meitu Xiuxiu.

We take regular interviews and all that, but in terms of using our tools to help the media agencies, I think they not particularly becauseright now how we organize our portfolio, we have a very industry specific strategy. So the first couple of apps is mostly focusing on the e-commerce industry.

Obviously, as a media agency, you can still use our app to create some of the banners or advertisements and so forth, but it's not geared towards that. So it's like you can use Photoshop to edit your face too, but you're better off using Meitu Xiuxiu.

Is that kind of analogy? It's easier that way, yeah.

So speaking of content creation, so Lin Yi was studying machine learning when he was in his graduate school and his video on Bilibili is talking about AI. So can you give us an example of the recent work you created with AI or with the help of AI?

Like my fight with AI interns,right? My recent contents are all somewhat related to AI tools. Like we are using it to generate illustrations and we use those illustrations in our videos. And definitely use ChatGPT or Claude to help with my process of writing my articles.

Yeah, that's basically my use case. And in this process, I was like, I really struggle with talking with AI agents. Their response is quick. Their attitude is good, but their capability is bad. What I think is currently the AI agents, like all those large language models, they have this problem of use depth.

It's actually not an original opinion from me. It's actually from the CEO of Minimax. I borrowed his idea. He said the current response of large language models are more like instinctive. It's like what you come to your mind like in one second.

And it's not a deeply thought answer. So that causes a very shallow use case. You can use AI agents, you can use large language models to do some simple tasks, but you cannot do complicated tasks. Like if I prepare for this session here, I have to go through many steps.

I need to look for some questions, I need to do some preparations, I need to see all those documents and generate some of my own ideas. But for large language models, that's just too much. And so basically they are still doing some of the additive work.

It's like the supplemental work is not fundamental work. The fundamental work is still through humans.

As Lin Yi was saying, I was really thinking about, okay, how do I turn that into how I really educate my students,right? Some of them are worrying about their jobs. They're still studying, so they need to start work looking for jobs in the next few months.

Some of them are already worrying about the jobs being gone. But I think what Lin Yi is kind of describing is that as a human being, you just need to get smarter and even more systematic and strategic in your thinking for you to actually be able to work with AI and have AI support you in the ways that's actually helpful,right?

But to do that, you actually need to really break down your tasks or your thinking process in a very systematic way. And only through that, then, and I think that's for individual level, but I think also for our industry level,right?

You need to break it down so that you know how which part of the workflow can AI actually step in and support. So I think that's something all of us need to do more thinking about.

The last part of this discussion, we are all consumers and users of information too. So I wanted to discuss about how we take control of what we read and what we consume, the information every day. So do you have any advice or, for example, where do you primarily get your information?

And is there anything you do deliberately to take control of the things you read and to tell which part AI has played in that piece of news?

If I want to do some formal study, I still go with formal journals and reports and papers. I'm working as a content creator, so I need to know what the community is talking about. So sometimes I have to dive into that pool of garbage and find some like what's our audience talking about.

So I talked about that previously. What I did is just lay back and see what the community or the algorithm, what they just feed me and take a look and get what people are talking about. My suggestion would be like, as well as what I'm doing is actually I will have some time, maybe like one day or two a week to just stay away from all of the content.

Yeah. Human brain have some limitation and in this time you just try to stay sane,right? Like in this age of information pollution here. Yeah, it's actually not a problem of artificial intelligence, I think. It's just a problem. Human always want to make some garbage.

In general, I still stick to traditional news and all that, but to actually learn something very deep, I still prefer to look at papers or talk to the expert. But I think what AI comes in very handy is that before you talk to an expert or like a CEO of another company or professor, you talk to GPT for a few moments first and it can actually get you up to speed on some of the relevant information, give you a little bit of concept so that you don't waste a lot of time when you are talking to the expert.

And granted, these large language models will have some errors in it, but that's where the expert comes in. He will say, okay, maybe this is what you get from GPT, but then this is not actually what is correct.

So I think as human beings, we have to find ways to let AI increase your efficiency, but you're not trying to rely on AI to sort of replace human expertise. I think we're still not quite there yet. Don't know what's going to happen in 10 years' time or maybe less.

Well, I'm glad to hear the old fashioned journalism still have some business here. As for myself, I get most of my information from social media, but I think I do have the privilege of most of my social media networks are from traditional journalists and media professionals.

So I kind of know that's reliable information instead of garbage. And one of my personal wish is actually I can build an AI agent to manage my information intake. I don't know how to do that. If any of you have ideas or where to start, I'm happy to get some inputs.

如果你认为有朋友也会喜欢本期 《 十字路口 》 的内容 ， 请转发微信推荐给他们 。 最后， 欢迎你加入 《 十字路口 》 的会员群 ， 我们会在群里每天放送 AI 全球新闻 ，并且鼓励大家在群里聊天互动交朋友 ， 寻找未来的同路人。