如何凭借联合创始人伊万·克雷科夫，让Buddy.ai成为全球领先的儿童AI家教

从学龄前到老年学习的创新，源于杰出的人才。

Innovation in preK to gray learning is powered by exceptional people.

十五多年来，各规模和各阶段的教育科技公司都信赖高等教育来发掘推动影响力的人才。

For over fifteen years, EdTech companies of all sizes and stages have trusted higher education to find the talent that drives impact.

当特定的技能和经验至关重要时，高等教育是能够提供支持的合作伙伴。

When specific skills and experiences are mission critical, higher education is a partner that delivers.

高等教育提供全职、兼职和高管招聘服务，深知您所需的市场人才。

Offering permanent, fractional, and executive recruitment, higher education knows the go to market talent you need.

了解更多，请访问 hireedu.com。

Learn more at hireedu.com.

那就是 hireedu.com。

That's hireedu.com.

所以我们走过的这条路是多年且相当昂贵的。

So the path we walked was a a multiyear and and pretty expensive.

因此，我们必须在多个国家、多种语言中推出我们的产品，收集大量数据，并训练我们的模型。

So we had to launch our product in multiple countries, in multiple languages, collect a lot of data, and train our model.

目前公开可用的儿童语音数据集最大也只有大约五百小时。

The largest publicly available dataset of kids speech is like five hundred hours.

这简直太差劲了。

It's just pathetic.

我的意思是，我们每个月收集的数据量是它的十倍，而这才是训练模型的唯一途径。

I mean, we collect 10 x of that every month, and that's the only way to train the model.

欢迎收听《教育科技内参》，这是覆盖教育科技行业的顶级播客。

Welcome to EdTech Insiders, the top podcast covering the education technology industry.

从融资轮次到影响力，再到早期教育、K12、高等教育和职场领域的AI进展，您在这里都能找到。

From funding rounds to impact to AI developments across early childhood, k 12, higher ed, and work, you'll find it all here at EdTech Insiders.

别忘了订阅我们的播客，查阅我们的简报，以及我们的活动日历。

Remember to subscribe to the pod, check out our newsletter, and also our event calendar.

如需深入了解，请访问《教育科技内参Plus》，您将能获取优质内容、加入我们的WhatsApp频道、优先参与活动，并获得来自亚历克斯和本的幕后洞察。

And to go deeper, check out EdTech Insiders Plus, where you can get premium content, access to our WhatsApp channel, early access to events, and back channel insights from Alex and Ben.

希望您喜欢今天的节目。

Hope you enjoyed today's pod.

今天我们邀请到了Ivan Crewkov，他是Buddy.dot.ai的首席执行官兼联合创始人，这是一家面向12岁以下儿童的领先AI家教平台，每年为全球超过两千万甚至更多的学生提供服务。

We're here with Ivan Crewkov, the CEO and cofounder of Buddy dot ai, the leading AI tutor for kids under 12, reaching over 20,000,000, even more, students annually from all over the world.

Buddy不仅仅是一个聊天机器人。

Buddy is not just a chatbot.

它是一个在移动应用中运行的、由AI驱动的会说话的动画角色，面向全球学生教授英语作为外语。

It's an AI powered talking animated character in a mobile app teaching English as a foreign language to students all over the world.

Buddy的独特之处在于，它在高度监管的环境中运行，而大多数现成的AI，比如ChatGPT，在这里无法使用。

Buddy is unique because it operates in a highly regulated environment where most off the shelf AIs like Chat GPT can't be used.

Ivan Crewkov，欢迎来到EdTech Insiders。

Ivan Crewkov, welcome to EdTech Insiders.

嗨，Alex。

Hey, Alex.

谢谢你的邀请。

Thank you for having me.

能来到这里，我感到非常荣幸和开心。

It's a pleasure and honor to be here.

我非常兴奋能和你交谈，因为我觉得你正在为人工智能学习的未来指明方向，而我很少对人这么说。

I'm so excited to talk to you because I feel like you're doing something that is really pointing the way towards the future of what AI learning will look like, and I don't say that that often to people.

所以首先，在我们深入讨论buddy.ai之前，能跟我们讲讲你自己吗？你是如何进入教育科技领域的？又是如何想到buddy.ai这个想法的？

So first off, before we get into what buddy.ai is, tell us a little bit about yourself and how you got into the education technology world and how you came up with this buddy.ai idea.

我是一名移民家长，同时也是一名语音AI开发者，这么说吧。

So I'm an immigrant parent and a voice AI developer, let's say.

这也就是我如何走到今天的。

And that's kinda how I got there.

我差不多十一年前带着我之前的创业项目搬到了美国。

I moved to The US almost eleven years ago with my previous startup.

在智能音箱这个品类还没被亚马逊Echo或谷歌Home等产品普及之前，我们就已经在开发智能音箱了，当时根本没有这些产品，我们真的是先驱。

We were building a smart speaker before this whole category was, like, created by so there was no Amazon Echoes or Google Homes, and we were thinking, like, we were real innovators.

我们当时准备发布我们的产品。

And we will launch our yeah.

我们正在筹备我们的Indiegogo众筹活动。

We were preparing to launch our Indiegogo campaign.

就在那之前两周，亚马逊发布了Echo，简直像一场灾难。

And two weeks before that, Amazon announced Echo, and it was like like a disaster.

但不管怎样，这是一次很棒的经历。

But, anyway, great experience.

那时候你学到了很多关于语音技术和对话式AI的知识。

You learned a lot about voice technology and conversational AI back then.

那是2014年，远在ChatGPT和Gemini出现之前。

It was 2014, long before Chat GPTs and Geminis.

是的。

And yeah.

于是我带着家人和我的女儿一起来了。

So I I brought my family with me and my my daughter.

她在学前班刚开始学英语的前四个月非常吃力，这对她来说是个大问题。

So she struggled to start speaking English in preschool for the first four months, and it was a real problem for her.

我们开始通过视频会议平台为她报名课程，因为那是真正教会孩子说话的唯一方式。

And we started taking lessons on platforms that connect children with live teachers via video conferencing because it was, like, the only way to actually teach a child to speak.

Duolingo很不错，但遗憾的是它不教口语。

Duolingo is great, but it does not teach speaking, unfortunately.

至少在那时候，它还不行。

At least back then, it wasn't.

于是我开始录制这些与真人教师的课程，并发现它们都是有脚本的。

And I just started recording those sessions with live teachers and realized that they were scripted.

教师们 literally 从屏幕上照着读，而且大多数人根本不是真正的教师，只是一些通过视频会议授课的普通人。

Like, teachers were literally reading from the their screens, and most of them, they weren't actual teachers, but some random people teaching via video conferencing.

我意识到，很可能80%的工作可以通过虚拟对话角色和语音识别来完成。

And I just realized that probably 80% of this work would be done using virtual talking eye character and speech recognition.

因此，我们可以极大地普及这个市场，因为当时价格极其昂贵。

And so we could, like, dramatically democratize this market because it was extremely expensive.

我知道即使在那时候，一节课也要15.20美元。

I know even back then, it was, like, $15.20 dollars per lesson.

所以我们的想法是，用一次真人辅导课的费用提供一个月的学习，这也就是我们走到今天的原因。

So the idea was to provide a month of learning at the cost of one live tutoring session, and that's kind of how we got to where we are.

是的

Yeah.

然后，当然，你最终打造的远远不止一个角色或简单的自动化教师。

And then, of course, what you came up with was beyond just a character or, you know, just a a sort of automated teacher.

这是一个完整的世界。

It's this whole world.

这是一个完整的Buddy AI学习伙伴。

It's this whole Buddy AI sort of learning companion.

你可以为他打扮。

You can dress him up.

你可以喂他吃饭。

He has you can feed him.

你可以和他一起冒险，你创造了一个完整的游戏化世界。

You have adventures with him, and you've sort of created this whole gamified world.

目前，关于Buddy.ai，我这里有两个不同的数据。

And at at this moment, Buddy.ai I have two different stats here.

告诉我哪个是正确的。

Tell me which one is right.

Buddy.ai 已经达到了五千万。

Buddy.ai has reached over 50,000,000.

对吗？

Is that right?

全球的学生？

Students globally?

现在快六千万了，亚历克斯。

And now now it's almost 60, Alex.

六千万。

60,000,000.

哇。

Wow.

所以你个人资料里写的是两千万。

So you you had 20,000,000 in your bio.

现在是五千万。

Now it's 50.

现在是六千万。

Now it's 60.

你们增长得非常快。

You are growing very quickly.

接近六千万名全球儿童。

60 almost 60,000,000 children globally.

告诉我们buddy.ai到底是什么，以及你们是如何开发出这个让孩子们如此兴奋的系统的。

Tell us about what buddy.ai actually is and how you developed this system that is so exciting for kids.

它很受欢迎，能创造出一种真正的沉浸式体验，我认为至少和与那些念稿的人类家教交谈一样令人兴奋，甚至可能更令人兴奋。

It's popular, and it just creates a real immersive experience that feels, I think, as at least as exciting, if not way more so, than talking to one of these human tutors that is reading a script.

是的，亚历克斯。

Yeah, Alex.

所以，如你所说，Buddy是一款AI教师。

So Buddy, as you said, is an AI teacher.

literally，它是一个会说话的动画虚拟角色，但我们特意将其设计为一个虚拟机器人，基本上就是一个卡通风格的机器人。

Like, literally, it's a it's a talking animated virtual character, but we purposely designed it as a virtual it's a robot, basically, like a cartoonish robot.

它的一个关键特性，我们称之为声音的魔力。

One of its, key properties, we call it magic of voice.

所以我们希望孩子们在与Buddy对话时，能感受到自己声音的力量。

So it's basically we want children to feel the power of their voice when they speak to Buddy.

每次你对Buddy说话，都像是在施一个咒语。

It's just like every time you say something to Buddy, it's like casting a spell.

我们为用户的发言提供了丰富的视觉反馈。

So we provide a rich visual feedback to what users are saying.

比如你说‘蓝色’，Buddy就会改变颜色。

You know, like, say blue and Buddy would change his color.

你说‘独角兽’，Buddy就会穿上独角兽的服装。

Say unicorn and Buddy would put on his unicorn costume.

类似这样的效果。

Things like that.

是的。

So yeah.

我们还希望创造一种类似宠物小精灵的体验，让你照顾你的虚拟朋友，就像玩游戏一样。

And we wanted also to create, you know, like a fill of a game in many ways close to, you know, like, Tamaguchi experiences where you take care of your virtual whatever friend path.

要做到这一点，你需要通过课程逐步进步。

And to do it, to do so, you need to progress through curriculum.

在这个过程中，你会遇到富有挑战性的教育体验并从中学习。

And that's where you face, you know, challenging educational experiences and learn.

你用语言描述它，但我真的建议任何听这个的人打开来看看Buddy实际长什么样。

You're describing it in words, but I really recommend anybody listening to this should open it up and look at what Buddy actually looks like.

他是一个非常可爱的角色。

He is an adorable character.

我 definitely 有这种类似宠物小精灵的感觉。

I definitely you have this sort of Tamagotchi.

它简直就像一只小小的飞行生物。

It's like almost like a little flying.

它脚上有火箭，还有一对大耳朵。

It has a rockets for feet and these big ears.

它可以穿上各种各样的衣服。

It can dress up in any kind of thing.

我懂Tamagotchi，但也懂Pokemon。

I get Tamagotchi, but I also get Pokemon.

我明白这看起来非常像一种高度复杂的儿童类产品，用行业术语来说。

I get a lot of these really it looks like a very, very sophisticated, you know, children's property, let's say, to use the sort of industry terms.

它能做很多不同的事情。

And it can do so many different things.

还有各种各样的小游戏。

And there are all of these different games.

有模式。

There's mode.

有一种战斗模式，看起来像小小的Pokemon模式。

There's sort of a fighting mode that looks like a little Pokemon mode.

还有这么多不同的装扮冒险。

There are all of these different dress up adventures.

所以在我们继续之前，能跟我们说说孩子们对这种学习方式的反应吗？

So before we even go further, tell us a little bit about how children react to this type of learning.

特别是，我很想了解一下，与任何传统学习方式相比如何。

Especially, I'd love to say, compared to sort of traditional learning of almost any kind.

当他们接触Buddy时，他们会怎么说？他们的反应是怎样的？

What do they say when they you know, how do they react to Buddy?

首先，这6000万次下载中，近70%是由孩子直接完成的。

So first of all, out of this, like, 60,000,000 downloads, almost 70% are done by children directly,

亚历克斯。

Alex.

虽然难以置信，但确实有四到五岁的孩子自己下载Buddy，然后教他们的父母使用，尤其是在美国以外的地区。

So it is hard to believe, but I don't know, like, four, five year old children download Buddy and then teach it to their parents, especially outside of The US.

在美国，主要是父母下载这款应用，但在其他地方，绝大多数都是孩子在主导。

So in The US where mostly parents download the app, but outside it's it's it's mostly all about children.

所以我们有一个完整的粉丝群体，YouTube 上有很多用户生成的关于 Buddy 的视频。

So we have, I don't know, a whole like fandom, a bunch of user generated videos on YouTube about Buddy.

因此，正如你所说，我们正试图打造一个 IP，构建一个游戏世界，真正为孩子们塑造一种印象——它不仅仅是一个应用或学习项目，而是一个独立的实体。

So we are trying to build an IP, as you said, like a a game world and really for children create an impression of not just an app or in, like, a learning program, but more as an entity.

他是一个外星探险机器人。

It's an alien explorer robot.

他有自己的 YouTube 频道，也在快速增长，我们的 YouTube 频道已经积累了超过两千万的自然观看量，诸如此类。

He has his YouTube channel, which is also growing, like, over 20,000,000 organic views on our YouTube channels, stuff like that.

是的。

Yeah.

你刚才顺口提到了 Duolingo，但我想说的是，我们接触过很多不同模式的教育科技公司。

You mentioned Duolingo sort of in passing there, but I think, you know, we talk to a lot of EdTech companies of all different types of models.

我总是觉得，直接面向学生和家长的教育科技公司，其运营模式和策略，与那些面向学校、学区和大学的公司截然不同。

And it it always strikes me as there's such a different playbook and such a different type of strategy for EdTech companies that are going directly to students and parents than one's going to schools and districts and colleges.

当我看到这里动画的精细程度、角色设计、服装搭配以及整体结构时，你非常清楚，你们是在和 YouTube 竞争。

And when I look at the sort of level of care put into the animations here, put into the character design, put into the costuming, put into the structure of it, you know very well that you're competing with YouTube.

你是在和迪士尼频道竞争。

You're competing with Disney Channel.

对吧？

Right?

你正在与其他许多高制作价值的儿童内容争夺孩子们的注意力。

You're competing with many other very high production value properties for kids' attention.

所以你在产品上投入了大量制作成本。

So you put a lot of production value into your product.

跟我们说说这些决策吧，因为我觉得我们在B2C领域经常看到这种做法，而且正如你所说，这种做法在B2C领域确实非常有效。

Tell us a little bit about those choices because I think this is something that we see a lot more on the b to c side, and I think it really pays off on the b to c side because, as you say, students find it.

孩子们会互相推荐。

They tell each other about it.

他们会直接下载。

They're gonna download it directly.

他们会非常喜欢玩它。

They're gonna love playing it.

但有时B2B公司根本不会这样思考。

But sometimes b two b companies just don't think like that at all.

我很想听听你谈谈你们做出的一些决策，比如：好吧。

I'd love to hear you just talk a little bit about sort of the decisions you make to say, okay.

我们真的希望这款产品看起来像孩子们在电视或YouTube上看到的任何内容一样精致、美观。

We really want this to look as polished, as beautiful as anything kids see on TV or on on YouTube.

是的。

Yeah.

亚历克斯，正如你提到的，我们正在与顶级手机游戏竞争。

Alex, as you mentioned, we are competing with top mobile games.

没错。

And Yes.

我们的目标是做到像顶级儿童手机游戏一样引人入胜。

We set a goal to be, you know, like, as engaging as top mobile games for kids.

我们选择的方式是组建一支团队，一边是AI工程师，另一边是实力强大的游戏设计团队。

The way we chose is actually we are building so our team combines, look, AI engineers on one side, but also a very strong game design team.

所以我们团队中有大量来自行业的专业人士。

So we are we are like a lot of we have a lot of people from industry.

比如我们的产品负责人是育碧的资深游戏爱好者。

Like, our head of product is a Ubisoft game love person.

我们的游戏设计负责人曾是《手术刀》和《装备七》的开发者。

We have a head of game design who is a former scalpel y king and outfit seven person.

我们的艺术总监也来自游戏行业。

Our art director is also from game industry.

因此，我们整个团队都是由游戏行业的专业人士组成，致力于让产品尽可能吸引人。

So we have a whole team of, like, game industry professionals working to make it as engaging as possible.

他们还与教育工作者合作。

And they're working with educators.

教育工作者主要设定学习目标。

So educators set basically, like, the objectives.

他们告诉我们，需要按这个顺序教授这些内容。

They told, we need to teach this and that and in that order.

然后我们的游戏设计团队会设计出游戏机制和所有那些酷炫的功能。

And then our game design team comes up with game mechanics and all of those cool things we have.

是的。

Yeah.

你们有这些高水平的游戏专家来制作游戏，但你们也有一个学校产品。

And then you have all of this very high production value game experts creating the games, but you do have a school product.

你们有一个类似Buddy的课堂版本。

You have a sort of classroom version of Buddy.

所以给我们讲讲它是怎么运作的。

So tell us how that works.

当你们向教师和课堂提供buddy.ai访问权限时，你们具体怎么做？又如何帮助他们取得成功？

How does the what do you do when you're giving teachers and classrooms access to buddy.ai, and how do you set them up for success?

亚历克斯，很遗憾，我们还在开发中。

Alex, unfortunately, we are still working on it.

事实上，许多教师今天已经在课堂上或一对一课程中使用Buddy，但并没有完整的功能支持。

So the fact is many teachers use Buddy today in their classrooms or in their one on one lessons without having a proper functionality.

不幸的是，我们还没有机会为教师专门开发这个功能。

And unfortunately, we haven't had an opportunity to develop this feature specifically for teachers.

所以这是一种非预期的使用方式，这意味着我们也能为教师和学校创造价值。

So it's like an unintended use, which means teachers we can create value for for teachers and schools as well.

目前，我们正在拉丁美洲的巴西和墨西哥与几所学校开展试点项目。

So we have several pilot projects now running with schools, mostly in Latin America, in Brazil, and Mexico.

但我们计划在明年大力推动这一方向。

But we plan, like, a real big push for next year.

因此，我们计划在明年进行下一轮融资，专门用于投资这些功能，让Buddy对教师更有用。

So we plan to raise our next round of funding, next year specifically to invest more in these features that will make Buddy even more useful for teachers.

报告、班级管理以及类似的功能。

The reporting, class management, and stuff like that

对。

Right.

作业布置。

Work assignment.

你们的课堂产品提供教师套餐许可，可以一次性购买20或30个许可证，并享受折扣。

Your classroom product, you offer, you know, teacher packs of licenses so you can get, you know, 20 or 30 licenses with with discounts.

显然，你们的产品已经被广泛用于许多课堂，尽管是非正式的使用方式。

You obviously have the ability to work, and you are being used in a lot of classrooms informally.

但正如你所说，这对教育科技公司来说始终是一个关键的决策点。

But as you mentioned, it's always an interesting decision point for for EdTech companies about this.

Duolingo 就是一个著名的例子，他们从未真正推出过面向学校的完整产品——当然，作为B2C教育科技的黄金标准，他们的产品在学校中被广泛使用。

And Duolingo famously really never did create a true school product that they're sort of, as you know, of course, the sort of gold standard of b to c EdTech, and they are used extensively in schools.

然而，他们从未真正开发出一套完整的学校版产品。

And yet they never actually created a full school product.

据我了解，他们并没有真正提供教师仪表盘、数据分析和管理控制等功能，而这些通常是其他公司会提供的。

And my understanding is they don't truly have, you know, teacher dashboards and analytics and admin controls and some of the things that people do.

这是一个有趣的选择，而你现在正站在十字路口，需要决定如何构建你们的教师端功能。

So it's an interesting choice, and you're obviously right at the sort of fork in the road there about how you're gonna build your teacher side.

我想象，在你们的战略会议室里，此刻正围绕着课堂使用场景进行着非常关键的讨论。

It's a very interesting moment, I can imagine, inside your strategy room thinking about what the classroom use case looks like.

你提到你们正在巴西开展业务，而Buddy另一个非常有趣且极具现代感的方面是，你们已经将Buddy提供给多种语言用户，包括巴西的葡萄牙语、波兰语、西班牙语、法语、俄语和土耳其语。

So you mentioned that you're working in Brazil, and one of the other things that's really interesting about Buddy and, of course, very modern about Edtech right now is that you have already have Buddy offered in many different languages, including Portuguese for Brazil, including Polish, including Spanish and French and Russian and Turkish.

请谈谈你们如何确保为不同语言背景的学生创造更多学习机会，尤其是你们教授的是英语。

Tell us about that process of making sure that you're creating more and more opportunities for different students with different linguistic backgrounds to learn, especially given that you're teaching English.

因此，来自不同语言背景的学生有多种途径进入你们的英语学习环境。

So you have so many different ways for students from different linguistic backgrounds to get into this English learning environment.

请谈谈你们是如何进行翻译的，以及如何利用技术和你们的背景加速支持更多语言。

Tell us about how you're doing that translation and how you're accelerating how you're using technology and your background to make sure that you have many different languages supported.

是的，亚历克斯。

Yeah, Alex.

首先，我们从第一天起就定位为一家国际公司，并且在进入美国市场之前，就已经在欧洲和拉丁美洲推出了产品。

So first of all, we actually started from day one as an international company, and we launched in Europe and Latin America actually before we have in The US.

我们做出这一选择是基于技术和监管方面的考虑。

And this choice we made because of technology and regulation.

我们之前已经讨论过，我们在COPPA及其他类似法规高度监管的领域运营。

So we already discussed that we operate in this highly regulated field with COPA and similar regulation.

在这个受严格监管的环境中，几乎没有现成的解决方案可以直接拿来搭建最小可行产品。

And so in this weird regulated world, there are not a lot of off the shelf solutions that you can just take and build an MVP.

所以我们不得不自己开发，当然，当时也有Soapbox Labs这样的解决方案，但它们价格昂贵，而且我们对质量并不满意。

So we had to build I mean, there were, of course, soapbox labs and solutions like that, but they were expensive, and we weren't happy with, with quality, actually.

因此，我们自主研发了一整套AI技术栈。

And so what we did, we built a whole AI stack in house.

而要做到这一点，你需要数据。

And to do it, you need data.

于是，我们选择在那些对12岁以下儿童数据收集没有严格监管的国家率先上线。

And so we just launched in countries where there were no strict regulation on data collection for children 12.

所以我们先在那里开始收集数据，训练我们的模型。

So we started collecting data there, teaching our our model.

当我们开发出的模型在儿童数据上的表现基本超越了谷歌时，才在美国上线，并获得了COPPA认证等资质。

And then when we had a model that basically outperformed Google on kids' data, then we launched in The US, got a COPPA certification, and, things like that.

这仅仅是我们的大型AI长期多阶段研发战略中的一部分。

So it was just a part of this, like, big AI long, multiyear AI building strategy that we had.

让我们深入探讨一下这一点。

Let's double click on that.

我刚才在问语言背景，但你回答的内容实际上更有趣，因为这正是当前教育科技领域最棘手的部分之一。

You know, I was asking about the language background, but what you're answering is actually even more interesting because this is such a tricky part of the EdTech landscape right now.

我们这个播客的听众都知道，Soapbox 曾是这个领域的 B2B 领先者，但后来被 Curriculum Associates 收购，变成了专有且封闭的系统。

Everybody our listeners to this podcast know that, you know, Soapbox was sort of the b to b lead in the space, but they got bought by curriculum associates and made sort of proprietary and private.

现在所有人都在拼命寻找方法，试图构建自己的儿童语音模型，或解决这个问题。

And everybody has been left sort of scrambling to figure out how to create their own kids' voice model or sort of to solve this.

听起来你们从很早就开始思考这个问题了。

It sounds like you've been thinking about it from very early on.

坦白说，去那些可以合法收集儿童语音数据的地方训练模型，而不必像在美国那样面对繁重的官僚程序，这真是非常聪明的策略。

And frankly, that's a very clever move to go to places where you can collect kids' voice data in a way that can train your model, but without having to go through quite as much of the bureaucratic red tape as happens in The US.

你们的模型在儿童语音识别上能超越谷歌，这非常令人印象深刻，我想这体现了你们团队卓越的技术实力。

The fact that you can outperform the Google model on kids for that is very impressive, and that, I think, is a tribute to the probably the technical prowess of your team.

请再多讲讲这个决策吧，因为我们已经访谈过几家试图做儿童语音识别的公司，它们都不得不面对各种挑战，尤其是自从 Soapbox 退出市场之后。

Tell us a little bit more about that decision because we've talked to a few different companies that are trying to do kids voice recognition, and they've all sort of had to wrestle with all the especially since Soapbox sort of came off the table.

你提到过，你在创办Buddy之前就已经在从事语音工程工作，专注于语音识别和智能音箱。

So you mentioned that you're a voice engineer, that you are already working on voice recognition and smart speakers before you started Buddy.

你认为教育科技领域应该如何应对这种共同的挑战，即需要为学生开发语音识别软件？

How would you recommend that the EdTech field sort of think about this collective problem of needing to have voice recognition software for students.

你显然已经构建了自己的专有模型。

You have obviously built your own proprietary model.

Soapbox 曾经向许多公司提供服务，但现在已完全整合进课程协会。

Soapbox was one that was offered to lots of different companies but is now fully incorporated into curriculum associates.

你认为整个行业该如何跨越这一障碍，确保孩子们能够用自己的声音操作这类AI工具？

How do you think the whole field is going to be able to get over this hump and make sure that kids can actually operate AI tools like this with their voices?

亚历克斯，这个问题真的很难回答。

Alex, it's it's such a hard question.

我知道。

I know.

我其实也不知道。

I actually don't know.

我们走的这条路是多年且非常昂贵的。

So the path we walked was a a multiyear and and pretty expensive.

所以我们不得不在多个国家、多种语言中推出我们的产品，收集大量数据，并训练我们的模型。

So we had to launch our product in multiple countries, in multiple languages, collect a lot of data, and train our model.

目前公开可用的儿童语音数据集最大也只有五百小时。

The largest publicly available data set of kids speech is like five hundred hours.

这简直太可怜了。

It's just pathetic.

我的意思是，我们每个月收集的数据量是它的十倍，这才是训练模型的唯一方法。

I mean, we collect 10 x of that every month, and that's the only way to train the model.

但现在除了Soapbox之外，还有其他供应商。

But there are other providers other to Sobox now.

我只是不确定它们有多好，因为我们一直专注于构建和训练我们自己的模型。

I just I'm I'm just unsure how good are they because we focused on building and training our own model.

我对当前市场上外包、开源以及其他替代方案的现状真的不太清楚。

I'm just unsure on this, like, the the status of outsourcing and, I mean, open source and other alternatives on the market today.

你们会考虑将你们的模型作为服务提供给其他教育科技公司吗？或者你觉得这个领域内的任何人会考虑这个选项吗？

Would you ever consider sort of offering your model as a service to other EdTech providers, or do you think anybody in the space is gonna consider that as an option?

是的。

Yeah.

我们一直在考虑这件事。

We've been thinking about it.

亚历克斯已经收到很多教育科技公司的询问。

Alex asked a lot by other EdTech companies.

所以我认为我们需要做一些工作，让这个模型真正具有市场吸引力，让任何人都能拿去并集成到他们的产品中。

So I think we will need to do some work to make it actually marketable so anybody could take it and plug it in their products.

但没错，在下一轮融资之后，我们肯定会考虑这个选项。

But, yeah, we'll we'll definitely consider this option after the next round of finding.

是的。

Yeah.

这非常有趣。

It's very interesting.

让我们更多地聊聊Buddy.ai中的学生体验，因为这非常有趣。

So let's talk a little bit more about the student experience in Buddy.ai because it's incredibly interesting.

你们真的是从零开始，完全以儿童为中心来构建的。

You really built it from the ground up with kids in mind.

里面的一切都是视频形式的。

Everything there's video.

还有游戏。

There's games.

有很多角色。

There's lots of characters.

还有很多内容，但你们同时也采用了严谨的课程体系，运用了间隔重复和教学法。

There's lots of but you also have the rigorous curriculum, and you use spaced repetition, and you use pedagogical methods.

你们是如何从零开始构建这个完整的英语学习世界的？又是如何利用AI来加速这一过程的？

How did you think about building this entire English language world from the ground up, and how did you use AI to really accelerate that process?

亚历克斯，首先，Buddy不仅仅是一个英语作为外语的产品。

Alex, so first of all, Buddy is not only English as a foreign language product already.

更大的理念是，打造一个AI导师平台，作为孩子的学习AI助手，从大约四岁开始陪伴孩子成长，逐步教授不同科目，同时收集学习模式数据，利用适应性学习技术制定个性化学习计划。

The bigger idea is, you know, like an AI tutor as a platform that a child's learning AI assistant that would grow with the child from, let's say, the age of four, teaching one subject after another while collecting learning pattern data and creating a personalized learning plan, like using adaptive learning technology.

这就是我们更大的愿景。

So that's the bigger vision we have.

目前，我们已有两个课程体系，第三个正在开发中。

So today, we have two curriculums and developing the third one.

最初的课程是英语作为外语教学，正如你提到的。

So original curriculum is in English as a foreign language, as you mentioned.

这是我们最大、最重要的产品。

That's our biggest, biggest product.

它贡献了我们80%以上的收入，用户占比甚至更高。

It's, like, 80% of our revenue for sure, but even more in in terms of users.

后来，我们推出了一个早期教育课程的最小可行版本。

Later, we launched, like, an early learning, like an MVP of an early learning course.

大约一年前，我们发现，在英语母语国家的用户中，约有一半人将Buddy用作语言治疗工具。

And about a year ago, we actually discovered that about half of our users in English speaking world have been using Buddy as a speech therapy tool.

我们的平台上有很多有特殊需求的孩子。

And we have a lot of children with special needs on the platform.

因此，我们开始开发这个新的专业语音治疗课程，我们知道，在英语国家，我们最活跃、最投入的用户就来自这个群体。

And so we started developing this new specialized speech therapy course, and we know that so I would say that in the English speaking world, so our most active and engaged users are in this space.

所以他们用Buddy进行语音和语言治疗。

So they use Buddy for speech therapy speech and language therapy.

是的。

Yeah.

这太惊人了。

That's wild.

这也体现了你们语音模型的出色表现。

And so and that's a tribute also to your voice model.

我不想再深入讨论，但人们在使用语音模型时常常遇到的问题是，它们必须能够适应各种口音和方言，还要能应对不同的语言障碍或各种说话方式。

Not to get back into it, but, I mean, some of the things that people wrestle with with voice models is that they have to be able to work with a wide variety of different accents and dialects, but also different speech impediments or various kinds of speaking.

你们的系统对需要语音治疗的学生如此有效，这确实令人振奋。

The fact that your system works that well for students who are in need of speech therapy actually works for them is incredibly exciting.

我喜欢这个方向。

I love that direction to go.

这是一个庞大的群体。

That's a big population.

这个群体真的、真的需要这种确切的支持。

It's a population that really, really, you know, needs support in this exact way.

学校里的职业治疗师和言语治疗师根本不够用，尤其是在全球范围内。

There are just simply not enough occupational and speech therapists, you know, out there in the schools to help, especially around the world.

我们马上回来。

We'll be right back.

Tuck Advisors 由曾创建并成功退出自己公司的企业家创立。

Tuck Advisors was founded by entrepreneurs who built and sold their own companies.

由于对其他并购公司感到不满，他们创建了自己理想中却找不到的那一家。

Frustrated by other m and a firms, they created the one they wished they could have hired but couldn't find.

一家真正理解创始人需求、以成交交易的百分比为北极星指标的公司。

One who understands what matters to founders and whose North Star KPI is the percentage of deals closed.

如果你正在考虑出售或收购一家教育科技公司，请立即联系Tuck Advisors。

If you're thinking of selling your EdTech company or buying one, contact Tuck Advisors now.

哦，我同意。

Oh, I agree.

而且我很自豪我们已经开始朝这个方向迈进。

And I'm super proud that we started moving in this direction.

而且这同时也是自然发生的。

And it also happened organically.

我们只是发现人们开始用我们的产品进行语言治疗。

We just learned that people started using our product for speech therapy.

所以当你谈到你关于这将成为一个AI学习伙伴的更广泛愿景时，这就是我第一次了解到Buddy的方式。

So when you talk about your broader vision about this being an AI learning companion, you know, that's how I first came across, Buddy.

当我们考察整个领域、了解谁在做什么时，你们进入了我们的视野，而在教育科技内幕市场地图的AI类别中，我们称之为学习伙伴。

You you came on our radar as we were looking across the landscape about who is doing what, and there's this category in the EdTech Insiders market map of AI we call learning companions.

但这个类别中的玩家实际上并没有我预期的那么多，至少目前是这样。

And there are not actually as many as I expected to be in there, at least so far.

我认为它们可能在增长。

I think they're probably growing.

这是一个不断发展的领域。

It's a growing category.

但你们在以非常深思熟虑的方式开展学习伴侣这一领域时，显然处于最前沿。

But you guys were very clearly at the very cutting front edge of doing learning companions in a really thoughtful way.

我的意思是，Buddy 在所有这些不同的体验和游戏中都表现得非常一致且持久。

I mean, Buddy really is just very consistent, persistent throughout all these different experiences and games.

我只能想象，人们会觉得他们真的了解他，还有那个袜子玩偶独角兽角色。

I can only imagine that people just feel like they know him and the the sock puppet unicorn character.

那个角色叫什么名字？

What is that character's name?

Sparkles。

Sparkles.

Sparkles。

Sparkles.

斯帕克莱斯先生。

Mister Sparkles.

没错。

Exactly.

他们太棒了。

They're fantastic.

我们在这里经常思考的一个问题，我相信你们也经常考虑，就是当你在设计一个持久的学习伙伴时，核心要素之一就是记忆——你对学生了解多少？

One thing that we wrestle with a lot here, and I'm sure you think about a lot too, is when you're thinking about a persistent learning companion, a core aspect of it is memory, is what do you know about the student?

他们喜欢做什么？

What do they like doing?

他们的语言问题是什么？

What do they what are their speech issues?

他们关心什么？

What do they care about?

他们对什么感到兴奋？

What are they excited about?

他们最喜欢的游戏是什么？

What's their favorite game?

我很想知道你是如何开始思考这个问题的——我肯定你已经深入思考过这一点。

I'm curious how you've begun to I'm sure you've thought a lot about this.

如果我们希望学生长期使用Buddy，几个月甚至几年，为了真正让体验尽可能成功和个性化，我们需要在长时间内了解这个学生哪些方面？

How you've begun to get into that idea of if we're gonna have students using Buddy for months or years on end, what do we have to know about that student over time to really make the experience as successful and personal as possible?

是的，亚历克斯。

Yeah, Alex.

如果目标是提供真正的个性化辅导，就像有一位真正的私人导师一样，那么我们就需要了解孩子的学习模式，而不仅仅是某些偏好。

So the if the idea is to provide actual personalized tutoring just like having an an an actual personal tutor, then we need to understand a child's learning patterns, you know, not just, like, some preferences.

比如说，这个角色——这个女孩——会更好。

But so let's say this character this girl better.

比如，这类游戏效果会更好。

This kind of games, for example, work better.

所以这个男孩需要更多的重复练习，诸如此类。

So this boy needs more repetition and things like that.

所以，要让这个系统长期有效，我们将使用自适应学习技术，比如基于人工智能的自适应学习。

And so, really, to make it work, like, longer term, we will use, like, adaptive learning, like, AI based adaptive learning technologies.

坦白说，我们才刚刚开始朝这个方向努力。

To be super frank, we just started working in this direction.

当然。

Sure.

到目前为止，我们主要专注于提供核心体验：语音识别，理解孩子说的话，并提供丰富、以对话驱动的教育性体验，目标明确但着眼于短期效果。

So far, we've been primarily focused on just providing core experience speech, understanding what children are saying and providing, like, a rich conversational voice driven experience that would be educational, so like real educational goals, but shorter term.

所以今天，我们的大多数学生都在六个月内完成我们现有的课程。

So today, most of our students, they just run through the curriculum that we have in under six months.

因此，这也是为什么我们目前平台上还没有年度学习计划的原因之一。

And so this is one of the things that and that's why we don't have, like, an annual plan, for example, today on the platform.

是的。

Yeah.

但我们正在不断添加越来越多的内容。

But, yeah, we are adding more and more and more content.

是的。

Yeah.

那就是是的。

That's a Yeah.

是的。

Yeah.

是的。

Yeah.

这是一个绝佳的机会，因为人们迅速完成了整个课程。

That's a great opportunity there where people are zipping through the entire curriculum.

这意味着你们有很大的扩展空间。

That means you have a lot of opportunity to expand.

这真的令人兴奋。

So that's really exciting.

是的。

Yes.

但孩子们即使学完了所有内容，仍然会继续和巴迪在一起，因为他们非常喜欢这个角色。

But children stick with Buddy even after they learn everything just because they they love the character.

我知道他们会给我们寄巴迪的照片，还会制作关于他的YouTube视频。

I know they send us pictures of of Buddy and produce YouTube videos about him.

所以，确实还有很大的改进空间。

So, yeah, there's definitely a huge room for improvement.

是的。

Yeah.

你说，这还处于初期阶段。

You say, this is nascent.

我认为整个领域都在研究这个问题。

I the whole field, I think, is working on this.

目前还没有人破解如何随着时间推移为每个学生建立知识库这个想法。

There's nobody has cracked this idea of how to build a knowledge base of in each individual student over time.

人们正在以各种方式研究这个问题，但我认为这是这一领域非常有趣的一个方面。

People are working on it in all different ways, but I think that's a really interesting aspect of this.

但让我们谈谈这个‘学习伙伴’的整体概念。

But let's talk just about this whole concept of learning companion.

就在我们今天连线时，我一直在思考一件事，那就是随着Anthropic、Google和OpenAI等公司纷纷在其核心产品中推出辅导模式和学习模式，这真的令人兴奋。

Right as we got on the the line today, something that has been that I've been thinking about a whole lot in the wake of all of these frontier model of Anthropic and Google and OpenAI all launching these tutoring modes, these sort of study modes inside their core products, which is really exciting.

我们在EdTech Insiders对此非常看好。

We're really very bullish on that, you know, at EdTech Insiders.

但它们所做的都是试图模拟一位导师、老师，或者至少看起来、感觉上和行为上有点像传统意义上的权威人物——与你一起学习的老师。

But what they are all doing is sort of trying to simulate a tutor or a teacher or, you know, somebody who looks and feels and acts at least somewhat like a traditional, what we'd consider, a authority figure, a teacher who's working with you.

我认为你们正在做一件非常具有前瞻性的、事实上尤其对年幼的孩子来说极其有力的事情，那就是你们在以一种不同的方式思考教学。

And I think something that you guys are doing that is incredibly forward thinking and, frankly, really, especially for younger kids, incredibly powerful is you're thinking about teaching in a different way.

它并不显得权威。

It doesn't feel authoritative.

它并不显得说教。

It doesn't feel didactic.

它让你感觉像在玩耍。

It feels like you're playing.

它让人感觉你是在与他合作、玩耍，帮助Buddy，给他打扮，和他一起玩游戏。

It feels like you're sort of collaborating and playing and helping Buddy out and dressing him up and and doing games alongside him.

我认为，这在未来的AI和教育科技领域将会越来越普遍，因为这是一条我们以前从未能触及的丰富路径，可以创造出真正富有魅力的AI角色，与学生并肩学习。

And it doesn't feel like I think that that is directionally something that we're going to see a lot more of in the future with AI and EdTech, because it just feels like a rich vein that we've literally never been able to do before to create really rich AI personalities that can work alongside students.

所以我非常想听听你对Buddy个性的看法，你们是如何让他成为一个孩子们愿意花时间陪伴、真正愿意向他学习的角色，而完全不像一个老师形象的。

So I'd love to hear how you think about sort of Buddy's personality and how you make him somebody who kids wanna spend time with and actually wanna learn from, and it doesn't feel like a sort of teacher personality at all.

是的。

Yeah.

对，亚历克斯。

Yeah, Alex.

谢谢。

Thank you.

这想法非常深刻。

That's super thoughtful.

很少有人能如此深入地思考。

Not a lot of people dig so deep.

但没错，从角色塑造的角度来看，Buddy更像一个学习伙伴。

But, yes, the idea is that from the character building standpoint, Buddy is more like a study buddy.

所以他和你一起学习。

So he learns with you.

他不会试图教你。

So he does not try to teach you.

我们故意从产品中去除了所有教育类词汇。

And we purposely remove all of the educational vocabulary from our product.

当Buddy与孩子互动时，他绝不会说‘学习’、‘掌握’这类词。

So, like, when Buddy interacts with a child, he would never say study, learn, or things like that.

他会说‘探索’、‘玩耍’、‘去冒险’，类似这样的表达。

He would say explore, play, go to adventure, and and things like that.

所以我们不用任何象征性的元素，比如虚拟教室之类的东西。

So we don't use any symbolic from a I don't like a virtual classrooms and stuff like that.

不用。

No.