AI 设计：捕捉魔力而不给用户增加负担

我认为这就是用 AI 进行设计的关键——你仍然需要捕捉它的本质和魔力，而不把负担转嫁给用户。想想那些现有的模式，但我们最初为什么要发明它们？那是因为我们有一些意图，并将其提炼成旋钮、搜索字段之类的东西。而现在，借助大语言模型和上下文理解，我们可以彻底重写这些模式。这是一个没有现成模式的模糊空间，你必须自己去摸索。

I think that's the thing that with designing with AI you need to still capture the essence and the magic of it without putting the burden on the user. Think about those existing patterns, but why did we invent them in the first place? It was because we had some intent that we distill down into knobs and search fields and stuff. And actually now with LLMs and contextual understanding, we can completely rewrite those patterns. This is an ambiguous space with no patterns, like you have to figure it out.

欢迎来到 Dive Club。我是 Rid，这里是设计师永不停歇学习的地方。本周的嘉宾是 Ammaar Reshi，他是 ElevenLabs 的设计主管，这个名字你可能听起来耳熟，因为它是之前在本节目中多次被提及的 AI 音频平台。所以这次对话我最喜欢的部分之一，就是听听设计 AI 原生产品是什么感觉，以及他们必须考虑的所有不同交互模式，还有那种从环境中诞生的实验文化。

Welcome to Dive Club. My name is Rid, and this is where designers never stop learning. This week's episode is with Ammaar Reshi, who's the head of design at ElevenLabs, and that might sound familiar because it's the AI audio platform that's been brought up multiple times on this show before. So one of my favorite parts of this conversation is just hearing what it's like designing an AI-native product and all of the different interaction patterns that they're having to think through, and also just the culture of experimentation that's been birthed out of that environment.

但在我们深入探讨 ElevenLabs 的设计工作细节之前，我请 Ammaar 给我们讲讲他早期一些 AI 实验背后的故事，因为其中有一个彻底火了。

But before we get into all of the details of how design works at ElevenLabs, I asked Ammaar to give us the backstory behind some of his earliest AI experiments, because one of them went completely viral.

我记得我告诉朋友们，我做过非常生动的梦，真希望有人能画出或重现其中的一些场景。那时候，我记得 OpenAI 演示了 DALL-E，如果你还记得的话，它当时需要排队，只有极少数人能访问，但看起来就像科幻小说。那时我开始四处寻找，心想：还有别的图像生成工具吗？我访问不了这个，但也许有其他东西，因为那正是我想要的梦境再现，对吧？我记得在 Twitter 的深处反复搜索，找到了一个当时大概只有 7000 粉丝的账号，叫 Midjourney。他们有一个研究预览版，你可以注册，然后他们会给你 Discord 的访问权限。当然，那大概是 Midjourney 的 v1 版本吧？我记得第一次输入提示词时，感觉太不真实了，你的想法仿佛活了过来。但 Midjourney 生成的人物看起来很奇怪，眼睛像有奇怪的洞之类的，所以你不能分享那些东西。我接近了梦境的氛围，但从未看到我想象中的场景。然后我就放下了。一开始我沉迷了大概三四天，不停地输入提示词，但后来觉得，它还没到那个程度。

I remember telling my friends actually that I had very, very vivid dreams and I wish I could have someone like paint or recreate a bunch of scenes from them. And it was about the time where I think OpenAI demoed DALL-E, if you remember it was super waitlisted and very few people had access, but it looked like science fiction. That's when I started looking around and I was like, is there another image generation thing out there? I don't have access to this, but maybe something else, because this is exactly what I want for those dreams, right? I remember searching and searching in the depths of Twitter and I found this account which probably had like 7,000 followers at the time called Midjourney, and they had a research preview you could sign up and they gave you access to the Discord. And of course that's Midjourney like v1 or something, right? And so I remember entering the prompt for the first time and I was just like, this is unreal, right? You just felt like your ideas came to life. But Midjourney wanted to, you know, people look really weird, eyes had like weird holes in them and stuff, so you couldn't share that stuff. And so I got close to the vibe of my dream but never like the scene that I wanted to see. And I left it. I was like, okay cool, at first I was addicted for maybe like 3 or 4 days just like prompting away, but then I was like, it's just not there yet.

快进到大概 2022 年 11 月，我一个朋友的女儿满 1 岁了，我去拜访他们。我经常和他们一起玩，还开玩笑说，嘿，我要当这个关系里的酷叔叔。有一天，她拉着我的手，带我上楼，我心想：怎么了？他说，她想让你给她读睡前故事。我当时想，哇，我终于成功了，你知道吗？我给她读这本书时，脑子里闪过一个念头：这本书太无聊了。他说，老兄，我每晚都得给她读这个，你知道吗？我说，好吧，嗯。那是一个旧金山的雨天周末，所以意味着要做点副业项目。我就在想，嗯，不知道 Midjourney 现在进步了多少。然后一个朋友告诉我 ChatGPT，说，嘿，这个新东西刚出来一周，看起来超棒。然后我灵光一闪：如果我把两者结合起来呢？我能为她做一本书吗？我给自己一个周末的时间来做。然后我想，我怎么出版呢？怎么拿到书？因为她不会读电子书，对吧？她需要实体书。我找到了亚马逊出版服务，填好了所有信息，24 小时内书就上架了，简直不可思议。然后我问，他们印平装本吗？他们印，3 天后我就拿到了一本平装书。

Fast forward to probably November 2022, a friend of mine, his daughter turned 1 year old, and I'd gone to visit them. I would always hang out with them and I joke like, hey, I'm going to be the cool uncle in this relationship. And one day she grabs my hand and she's walking me upstairs, and I'm like, what's happening? And he's like, she wants you to read her a bedtime story. And I was like, whoa, I'm finally making it, you know? And I'm reading her this book and the thought crossing my mind was like, this is so boring, this book. And he was like, dude, I have to read this to her every night, you know? And I was like, okay, hmm. And it was a rainy weekend in SF, so that means side project. So I basically was like, huh, I wonder like Midjourney, I wonder how much better it's gotten. And then a friend of mine was telling me about ChatGPT and was like, hey, this new thing's come out, it's a week old, it looks crazy. And then it just clicked in my head. I was like, what if I combine the two? Could I make a book for her? And as I was making it, I gave myself just the weekend to do it. And then I was like, well, how do I publish this? How do I get this? Because she's not going to read it like a digital book, right? She needs something physical. And I found Amazon publishing and I just filled it all out, and within 24 hours the book was on the Amazon store, which was unreal. And then I was like, do they print paperback? And they do, and I had a paperback in my hand in 3 days.

我记得我读到那条推文时在哪里，因为我跑过去给我妻子看，我说，看看这家伙做了什么，我简直不敢相信。那时我意识到，我们可以创作整个系列的儿童书，也许孩子消费的很多东西，你都可以和他们一起共同创作。这完全颠覆了我对可能性的认知。

I remember where I was when I read the tweet, because I ran and I showed it to my wife and I'm like, look what this dude did, I can't believe this. And it was like dawning on me that we could create whole series of kids books and like maybe a lot of what your kid would consume you could actually co-create together. And it totally blew the roof off of what I thought was possible.

这正是我收到的那些信息。有人说，天哪，我和我女儿一直在编故事，现在我迫不及待想和她一起真正做一本。另一个人说，我 8 岁的孩子写了很多故事，现在他可以去给它们配插图了。所以看到人们一直在寻找这样的东西，这真是太好了，对很多人来说，这就像是一个顿悟时刻，看到这些真的很有趣。

And that was exactly the kinds of messages I was getting. There were people who were like, oh my god, I've been me and my daughter have been like crafting up stories and now I can't wait to like actually make one with her. Or another person had said, my 8-year-old writes a ton of stories and now he can actually go and illustrate them. So it was really cool to see that people had already been looking for something like this and this was kind of the aha moment for a lot of folks, which was yeah, it was really fun to see that.

快速插播一条消息，然后我们再回到正题。你知道我一直在想什么吗？就是设计师版的 Cursor。这就是为什么我全力支持 Deson。它不是另一个从零到一的原型或副业工具。它是一个可视化界面，让你可以访问公司现有的代码库，从而更新组件、排版、属性等内容，并将你的设计推送到生产环境，而无需经过工程师。这就是产品设计的未来。你今天就可以开始体验，只需访问 dive.club/deson 即可开始。拼写是 D-E-S-O-N。

Real quick message and then we can jump back into it. You know what I can't stop thinking about? Cursor but for designers. And that's why I'm all in on Deson. It's not another zero-to-one tool for prototypes or side projects. It's a visual interface that gives you access to your company's existing codebase so you can make updates to things like components, typography, properties, and push your designs to production without having to go through an engineer. And that's what the future of product design will feel like. And you can start experiencing it today, just head to dive.club/deson to get started. That's D-E-S-O-N.

听着，你知道我是 Jitter 的超级粉丝，但他们最新的版本绝对是迄今为止最好的。现在 Jitter 有了一个用于动画设计的无限画布，我的意思是，使用它 30 秒内我就知道，这本来就是动态设计应有的方式。它让迭代变得容易 10 倍，与团队成员协作也更有意义，而且你可以通过单个文件中的多个画板跨格式缩放内容。Jitter 太棒了，这个产品好得令人难以置信。你现在就得试试，只需访问 dive。

Look, you know how big of a Jitter fan I am, but their latest release is hands down the best one yet. Now Jitter has an infinite canvas for animation design, and I mean within 30 seconds of using it I knew that this was always the way that it should be for motion design. It makes iteration 10 times easier, collaborating with team members makes way more sense, and you can scale content across formats using multiple artboards in a single file. Jitter is crazy, like this product is so incredibly good. Now you got to try it out, just head to dive.

请多谈谈你的个人实验之旅，因为这确实开启了一系列让我深受启发的项目。

Take us down your personal experiment journey a little bit more, because this really kicked off a whole series that I find so inspiring.

是的，谢谢。我脑子里突然意识到，如果你能想到并想象出某样东西，很可能就有一个 AI 工具能帮你实现它。这就是我们当时所处的浪潮，对吧？ChatGPT 时代，人们用提示词写书、写文章，做各种事情。学校都慌了，对吧？然后视频生成器出现了，突然间你可以让想法动起来。图像生成器也变得更好了。所以对我来说，这是一个巨大的创意出口，因为就像我说的，这又回到了那个梦想的事情。我想，现在我终于可以讲述我一直想讲的故事了，这能弥补我确实不擅长掌握的技能。它让我提升了一个层次。直到今天我还是这么看。它是你技能的增强，真正把你的想法提升到新高度。所以对我来说，就像，好吧，还有哪些有趣的创意项目或实验我想做？那就是，我想看看是什么感觉，你知道那个常被问到的问题：如果你能和已故或在世的人共进晚餐，你会选谁？我就想，如果我能和过去的人一起做个播客呢？有没有工具能让他们的声音复活？ElevenLabs，对吧？所以那就是一个项目。然后我学过计算机科学，但太理论化了，我从不觉得自己有从零到一构建应用的实际技能。然后，你知道，编码随着 GPT-4 变得好多了。我记得我给自己设定的那个任务，真是痛苦。但无知是福，真的。所以，我小时候玩过一个 DOS 游戏叫 SkyRoads，我就想，如果我能用 GPT-4 重制它，做成 3D 的呢？说实话，用 GPT-4 我再也不会这么干了，但我整个周末都在搞，用了两个 OpenAI 账户，付了订阅费限制，直到成功。我做出了游戏，它运行了。这对我来说更是证明了，只要努力，一切皆有可能。我认为很重要的一点是：不要从工具的角度看，只管你想做什么，然后开始寻找能实现它的东西。有没有办法生成图像？有没有办法编写视频游戏或从游戏中提取模型？还有，我总跟人说：如果这次没成功，先保留那个想法，因为两三个月或六个月后它就会实现。自从这些模型变得更好以来，情况一直如此。

Yeah, thank you. It basically clicked in my head that if you could think of something and imagine it, then there's likely an AI tool that'll help you make it. And that's kind of the wave we were riding at that time, right? ChatGPT, people were prompting books, people were prompting essays and all sorts of things. Schools were freaking out, right? And then video generators came out, and suddenly you could visualize your ideas in motion. Image generators got better. So for me, it was a massive creative outlet, because like I was saying, it goes back to the whole dream thing. I was like, well, now I can finally tell the stories I've been wanting to tell, and this can patch the skills that I honestly just wasn't good at acquiring. And it kind of leveled me up. That's how I view it today still. It's an augmentation of your skills and really takes your ideas to the next level. So for me, it was like, okay, what are all the fun other creative projects or experiments I would want to do? And that was, I wanted to see what it was like, you know, that question you always get: if you could have dinner with someone dead or alive, who would it be? And I was like, what if I could just make a podcast with someone from the past? And are there tools that allow me to bring their voice to life? ElevenLabs, right? So that was that one. And then I'd studied computer science, but it was so theoretical that I never felt like I had the practical skills to actually build an app from zero to one. And then, you know, coding got so much better with GPT-4. And I remember the task I gave myself with that one, and it was excruciating. But ignorance is bliss, honestly. And so, there was this video game I used to play as a kid, a DOS game called SkyRoads, and I was like, what if I could recreate that with GPT-4, make it 3D? Which honestly, with GPT-4, I would just never do again, but I just went at it a whole weekend, like two OpenAI accounts, paying for that subscription limit, until it happened. And I made the game, and it worked. And that for me was just even more affirmation of this anything is possible if you just push. And I think it's really important: don't look at it from the tool perspective, just go for what do you want to make, and then start looking for things that can make that happen. Are there ways to generate images? Is there a way to code a video game or pull models from a video game? And also, I always tell people this: if it didn't work this time, save that idea because it's going to work in two to three months or six months from now. And that's been the case since these models have gotten so much better.

我自己都有点笑了，因为你用“痛苦”来形容 GPT-4，而当时它可能感觉像魔法。就像，我们回头看今天的世界，会觉得，天哪，那真是痛苦，但今天它感觉像魔法。

I was kind of laughing myself because you said the word 'excruciating' about GPT-4, which probably felt like magic at the time. And it's like, we're going to look back at the current state of the world today and like, man, that was excruciating, but it feels like magic today though.

没错，太对了。GPT-4 刚出来时简直不真实。就像，终于我能完全编程了。所以对我来说，让我持续前进的反馈循环是，我在 Twitter 上完全不加过滤地分享我的想法，不是那种为了互动而钓鱼，纯粹就是，这是我做的，这是我是怎么做的。然后有一个很棒的循环：要么别人重新混搭这个想法，要么他们想知道我是怎么做的，或者给我看他们做的东西。这让我一直前进，因为我喜欢那种兴奋感，喜欢分享知识，就这样不断投入，不断前进。

No, it's so true. GPT-4 was unreal when it came out. This was like, finally I'll be able to fully code. And so I think for me, the feedback loop that kept me going was, I kept sharing my ideas on Twitter completely unfiltered, not like engagement baiting or anything like that, purely just like, here's what I did, here's how I did it. And there was such a great loop of either people remixing the idea or people wanting to know how I did it, or showing me something they made. And that just kept me going because I loved the excitement, I loved sharing the knowledge and just kind of feeding into that and going and going.

我想快速谈谈 Twitter 这部分，因为这正是我们能有这次对话的原因。我最初就是这样发现你的书的。在此之前你在 Twitter 上活跃吗？因为至少从我的外部视角看，这是一个相当大的上升，而实际上你只是持续出现并分享你的实验。

I want to talk about the Twitter piece actually really quickly, because that is like why we're even having this conversation. That's how I found your book in the beginning. How much were you even doing on Twitter before this? Because at least from my outside perspective, it's been quite the ascent, and really all it has been is just you continually showing up and sharing your experiments.

是的，说得好。我之前其实在 Twitter 上并不活跃。我会关注别人，只是潜水，但从不真正发帖。我有一个朋友，当我做出那本书时，我对那个体验感到非常震撼。我记得我把它发到 Instagram 上，分享给朋友们，我说，嘿，如果有人想要一本，我可以寄给你。我太兴奋了，在这本书上亏了一大笔钱，到处寄送。我朋友说，你一定要在 Twitter 上发这个，这个过程太酷了。然后我按实际步骤写了那个帖子。头 12 个小时，它得到了大约 100 个赞。那对我来说是前所未有的。所以我想，这太棒了，哇，科技 Twitter 喜欢这个。然后我记得凌晨四五点醒来，手机整晚都在嗡嗡响。我想，怎么回事？我是不是不小心设了个凌晨 5 点的闹钟？

Yeah, it's a great point. I actually was not that active on Twitter prior. I would, you know, follow around and just like lurk, but never really post. And a friend of mine, when I made the book, I was just so blown away by the experience. I remember posting it on my Instagram and sharing with my friends, and I said, hey, if anyone wants a copy, I'll send it to you. I was so excited, lost a ton of money on this book, shipping it everywhere. One of my friends is like, you must post about this on Twitter, this is such a cool process. And I wrote the thread exactly how I did it. And in the first like 12 hours, it got like 100-ish likes. And that for me was like the most I'd ever had. And so I was like, this is amazing, like wow, Tech Twitter loves this. And then I remember waking up at 4 or 5 a.m. and my phone is just like buzzing away off the night. I was like, what's going on? Did I set an alarm by accident for 5:00 a.m.?

而且看起来，简直没完没了。这就是事情变得复杂的地方。那是来自艺术家、作家的恐惧。这条推文从 100 个赞，大概 5 万次观看，一下子涨到了 200 万次，最终达到 1100 万次。然后你醒来发现《时代》杂志、BBC 和《华盛顿邮报》都在问：“你怎么看 AI 和艺术，以及它们的未来？” 你心想：“好吧，看来我得认真想想我刚刚做的事情到底意味着什么。” 但这真是一段奇妙的旅程。

And it look, and it's just endless. And that's where it got really mixed. It was the fear from artists, from writers. This tweet had crossed from 100 likes, I think it was like 50k views, to at that point 2 million views, eventually 11 million. And then you're waking up to like Time Magazine and BBC and Washington Post, and they're like, 'What do you think about AI and art and the future of that?' And you're like, 'Okay, I guess I have to think a little hard about what the implications of what I just did are.' But it was an amazing, amazing journey.

跟我聊聊 ElevenLabs 是怎么进入你的视野的。你在 Brex 做设计总监，那是个很不同的领域。这段经历是怎样的？

Talk to me a little bit about how ElevenLabs came into the picture. So you're a design director at Brex, pretty different space. What was that journey like?

嗯，好的，先简单说说 Brex：Brex 也是一段很棒的旅程。我是在 2021 年还是 2020 年加入的，当时正好是金融科技热潮。那里正在组建一个很棒的团队。我在 Palantir 的前设计主管刚加入，我很喜欢和他共事。所以感觉是加入一家高速增长初创公司的好时机。我很庆幸自己这么做了，因为我觉得这为我在 11 的工作做好了准备。那也是高速增长。我们的设计团队从最初的四五个人，一度增长到 50 人。公司从 200 人增长到 1500 人。当你看到这种情况时，专注是关键。随着每个团队都努力向前推进，保持质量门槛变得极具挑战性，但你仍然需要有愿景，并主动推进。所以你一边快速扩张，一边要满足每个工程师的需求。这中间有一个巨大的负载平衡挑战。有了那段经历，最后我管理着不少设计师和整个团队组织，我真的很享受那个挑战。那是全新的、不同的。而 AI 这件事，老实说只是我业余时间为了好玩而做的事，周末发发帖子。但对我来说，那是创意的解锁，因为坦率地说，金融科技没那么性感。作为一个喜欢创意和讲故事的人，这就像打开了那扇门。有一次 Brex 有点担心，说：“你在 AI 项目上做了很多副业，这会影响工作吗？” 我实际上回答说：“这其实对工作有帮助，因为这种创意解锁给了我更多的能量和设计创意，我会带回工作中。” 这也是我会鼓励其他设计师去做的。我们每天用 Figma，然后你会说：“我没时间做自己的事，周末再也不想看到 Figma 了。” 但如果你找到能表达想法并获得反馈循环的东西，那会是巨大的推动。我也有这种感觉，比如周末我会试着做点小东西，虽然很傻，但创造带来的能量是我多年来从未体验过的——一切都是那么新鲜，你感觉自己又像个孩子，抓起笔记本电脑到处跑，给每个人看：“看我做了什么！” 我很久没有过这种感觉了，而且我确实认为这种感觉会反哺到日常工作，或者随便你怎么称呼它。

Yeah, well, so on Brex real quick: Brex was an amazing ride too. I think it was 2021, 2020 when I joined, and that was kind of the fintech boom taking place around that time. An amazing team was forming there. My former head of design at Palantir just joined, and I loved working with them. So it felt like a great period to join a hypergrowth startup. And I'm so glad I did, because I think it really prepared me for what 11 was also going to be. It was hypergrowth. Our design team grew from the early four or five of us to, I think, 50 at one point. The company grew from 200 to 1500. And when you see that, it's like focus is so key. Maintaining the quality bar becomes incredibly challenging as every team tries to push forward, but you still need to have a vision and push ahead and be proactive. So you're scaling really fast and meeting the demands of every engineer. There's a huge load balancing challenge there. Having had that experience, and then finally to the point where I was managing quite a few designers and a whole team and org, I was really enjoying that challenge. It was completely new and different. And this AI thing was honestly just a thing I was doing on the side for fun and posting on my weekends. But that was for me the creative unlock, when frankly, fintech isn't that sexy. So as someone who just enjoys being creative and storytelling, this was the unlock for that. There was a point where Brex was kind of worried, like, 'You're doing a lot of stuff on the side with your AI projects. How does that affect work?' And I actually was like, 'It's actually helping work, because this has been a creative unlock that's given me way more energy and design creativity that I bring back to work.' And that's what I would encourage other designers to do too. I think we do Figma on the daily, and then you're like, 'I don't have time for my own things, I don't want to see Figma again on the weekend.' But if you find something that you can put your ideas out with and have that feedback loop, it's such a boost. I'm feeling that a little bit like just over the weekends I try to just make a little thing, and it's so silly, but there is an energy about creating that I haven't experienced in years, actually, where everything's so new and so fresh, and you feel like a kid again, just grabbing your laptop and running around and showing it to everyone like, 'Look what I made!' I've been having that moment for the first time in a long time, and I do think it really does trickle back into the day job, or however you want to refer to it.

你说得对，五年前的编程教程是这样的：“这是如何写你的第一个 Hello World。” 你会给多少人看？他们只会说：“哦，不错。” 说得太好了。而现在呢，“嘿，我刚做了贪吃蛇和乒乓球，是个多人游戏，戴上你的 VR 头显，你就是那个球。” 这完全改变了我们能分享的东西。当你从朋友那里得到那种兴奋的反馈循环时，它会极大地激励你想做更多。

And to your point, the programming tutorials of like five years ago were, 'Here's how you write your first Hello World.' And how many people are you going to show that to? They're just going to be like, 'Okay, cool.' Such a good point. And now it's like, 'Hey, I just made Snake and Pong, and it's a multiplayer game, and put on your VR headset and you'll be the ball.' It's completely changed what we can share. And when you get that excitement feedback loop from your friends, it's a phenomenal boost in wanting you to do more.

是的，我认为这是很重要的一部分。在 Brex 工作并做这些副项目时，其中一个最初是一个播客。那天是总统日，所以我想：“哦，不知道能不能和过去的总统聊聊天？那会是什么感觉？” 那是个很有趣的实验。我选了罗纳德·里根，因为我觉得在当前政治气候下，很难选出一个最有争议的人。我选了里根，然后让 Bing Chat 浏览他的历史，假设他会如何回答我写下的一系列问题。然后我想：“有没有什么工具可以重现他的声音？” ElevenLabs，不过现在出于明显的原因已经不允许了。但没错，我做了那个实验，和他进行了一次完整的播客对话。很多人说：“我想要一个系列。你下次能和这个人聊吗？再下次和那个人聊？” 但从那时起，我开始把 ElevenLabs 看作一个非常有趣的音频叙事工具。很快它变成了一个平台。他们有有声书，可以用各种声音完整叙述。然后我想：“好吧，酷。我喜欢制作 AI 视频，现在终于可以用这些不同的声音给角色配音了。” 所以我继续用 ElevenLabs 做东西。后来有一次，我花了很多积分，就联系了一个朋友问：“你认识 ElevenLabs 的人吗？” 她说：“其实，让我介绍你给创始人认识。他很乐意听听你的项目。” 我们聊得很愉快，他说：“嘿，项目的事先放一边，我们会给你积分。我其实在找一个人来领导设计，并弥合工程端和讲述技术可能性之间的差距。” 我之前没考虑过，因为我当时确实还喜欢 Brex。但后来我见了团队，吸引力太大了。而且 Mati 非常擅长促成合作。所以我非常兴奋。是的，从那以后就是一段疯狂的旅程了。

Yeah, I think that's a huge part of it. As I was doing Brex and making all these side projects, one of those was first a podcast. It was on President's Day, so I was like, 'Oh, I wonder if I could chat with a president from the past. How would that be?' It was a really fun experiment. I went for Ronald Reagan because I thought it was hard to pick who would be the most polarizing given the current political climate. I went to Ronald Reagan, and I asked Bing Chat to just kind of go through his history and assume how he would react to a series of questions I had written down. Then I was like, 'Well, is there anything out there that I could use to recreate his voice?' ElevenLabs, which is not allowed anymore for very obvious reasons. But yeah, I did that experiment, and I had this podcast conversation going with him. It was a full-on conversation. And a lot of people were like, 'I want a series. Can you chat with this person next and that person next?' But that's when I started to see ElevenLabs as this really fun storytelling tool from an audio perspective. And soon it was becoming a platform. They had audiobooks that you could have fully narrated with voices. Then I was like, 'Okay, cool. I love making AI videos, and now I can finally voice the characters with all these other voices.' So I just kept making stuff with ElevenLabs. Then at one point, I was spending so much on credits that I reached out to a friend and was like, 'Do you know anyone at ElevenLabs?' And she was like, 'Actually, let me introduce you to the founder. He'd love to hear about your project.' We had a great chat, and he was saying, 'Hey, project aside, we'll give you the credits. I'm actually looking for someone to kind of lead design and bridge that gap between the engineering side and telling the story of what's possible with the technology.' I hadn't considered it because I was still honestly enjoying Brex. Then I met the team, and the draw was too good. And Mati is very good at closing. So I was pumped. Yeah, it's been a wild ride since.

那么，为了让你了解完整的背景，你加入时团队的状态是怎样的？也许你可以给那些在节目中听说过 ElevenLabs、但实际并不清楚界面或软件能实现什么的人简单介绍一下。你能先给 ElevenLabs 一点背景，然后我们再深入探讨吗？

So just to even get the full context, what was the state of the team when you joined? And maybe you could even just give a little bit of a rundown for people who have heard ElevenLabs on this show but don't actually have like an interface in mind or what the software even is making possible for people. Can you just give it a little context for ElevenLabs and then we can kind of drill in from there?

好的。是的，ElevenLabs 是一个 AI 音频平台。公司的核心使命是让内容在任何语言、任何声音下都可访问。我们正在构建音频 AI 的基础设施。核心产品是语音合成，你可以从一小段样本中克隆声音，并用那个声音生成语音。我们还有音效生成，并且正在开发音乐。我们的想法是给创作者和开发者提供工具，让他们能制作出听起来自然且富有表现力的音频内容。我加入时，团队很小，大概 20 人左右。主要是工程人员，非常专注于模型和 API。设计基本上就我一个人。所以我必须从零开始建立设计职能，这很令人兴奋。产品已经被很多人使用，但界面非常简陋。我的工作是让它更易用，并讲述技术可能性的故事。

Cool. Yeah, so ElevenLabs is an AI audio platform. The main thing here and the mission of the company is to make content accessible in any language, in any voice. So we're building the infrastructure for audio AI. The core product is voice synthesis, where you can clone a voice from a short sample and generate speech in that voice. We also have sound effects generation, and we're working on music. The idea is to give creators and developers the tools to create audio content that sounds natural and expressive. When I joined, the team was small, maybe around 20 people. It was very engineering-heavy, with a strong focus on the model and the API. Design was essentially me. So I had to build the design function from scratch, which was exciting. The product was already being used by a lot of people, but the interface was very bare-bones. My job was to make it more accessible and tell the story of what's possible.

让音频在任意语言和任意声音中完全通用可访问。这意味着我们可以在保持原说话者真实性的同时为视频配音，但你能听到西班牙口音或德国口音等，从而获得更好的配音效果。我们有语音生成功能，可以复制你自己的声音。我们还有一个平台，出色的配音演员可以借出他们的声音并从中获利，我们可以生成各种非常逼真的文本转语音。这已被用于视频画外音、无脸 YouTube 频道，或者如果你在制作卡通片并想为角色配音、视频游戏角色等，用途广泛。甚至在 CNN 上，特朗普审判的旁白也使用了 ElevenLabs 的声音。你看到各种各样的应用。我最近最喜欢的一个案例是，很多不幸患有 ALS 并失去声音的人，ElevenLabs 让他们重新获得了声音。他们复制了声音，能够与家人交流、发表演讲。看到声音方面的各种应用真是太棒了。当然，我们也做音效，很快会做音乐，然后是整个音频领域。这就是 ElevenLabs 的概况。我加入时团队大约有 30 人，当时只有一名设计承包商，主要是工程和研究。那是一个非常早期的阶段。我认为 B 轮融资即将宣布，品牌也在进行我们正在处理的更新。所以很多基础性工作，非常早期。

Make audio completely universally accessible in any language and any voice. That means we can dub videos while maintaining the authenticity of the original speaker, but you can hear them in a Spanish accent or a German accent or whatever, which means way better dubs. You have audio generation for voices, so you can replicate your own voice. We also have a platform where amazing voice actors lend their voice, earn from it, and we can generate all sorts of text-to-speech with very realistic sounding voices. That's been used for video voiceovers, faceless YouTube channels, or if you're making a cartoon and you want to voice your character, video game characters, there's a whole range. Even on CNN, the Trump trials were narrated with ElevenLabs voices. You're getting to see a whole range. My favorite one very recently has been a lot of people who unfortunately suffer from ALS and lose their voice. ElevenLabs has given them their voice back. They've replicated their voice, they've been able to communicate with family, deliver speeches. It's been amazing to see the range of stuff just on the voice front. And then of course we also do sound effects, soon music, then the whole gamut of audio. So that's the rundown of ElevenLabs. The state of the team when I joined was about 30ish people, one design contractor at the time, and primarily engineering and research. It was a very early stage. I think the Series B was just about to be announced, and the brand was going through a refresh that we were working through. So a lot of foundational stuff, very early on in the mix.

我想了解更多在这种环境和这类产品中工作的感受。也许从高层次开始：设计一个没有太多先例的未来是什么感觉？你们在很多方面都引领着潮流。作为设计师，那是什么感觉？

I want to learn a bunch more about just what it's like working in this kind of environment and on this type of product. So maybe even just starting high level: what's it like designing the future of something where there's just not as much precedent? You all are leading the charge in so many ways. What's that like as a designer?

这是一个非常有趣的挑战，也非常有回报，因为正如你所说，没有先例，你可以看到自己的模式成为被复制的模式。但你也可能会想，我不知道那个模式是否正确，所以你也可以通过那种方法在那种规模上进行测试。前几天我和浏览器公司的 Nate Parrot 聊天，我们来回辩论。我说，提示框只是一个文本字段和一个生成按钮，这真的是我们能做的最好的吗？他的观点实际上是，我们将大量复杂性抽象为简单的聊天，这也许就是设计上的举措。也许它确实需要这么简单。但我也认为这里还有一步前进，那就是这些模型可以做很多事情。你可以调整各种参数，改变模型类型，当然每个模型要么推理速度更快，要么能更好地控制稳定性等等。但你并不希望人们去担心这些事情。所以即使我在考虑设计音乐时，你可以添加歌词，改变流派，也许可以调整曲目和特定片段的氛围部分。但最自然的方式是什么？最初我可能只从文本开始，你描述每一个方面。但这需要你的用户知道所有这些东西，才能写出一个好的提示来获得好结果。我开始更多地考虑一种类似动态岛的提示框，如果你需要更改歌词，它会变形为一个提示框，基本上围绕歌词如何编辑和歌词部分进行最佳设计。如果你要选择流派，它不会让你思考，而是填充所有你可能想尝试或考虑的流派标签。你可以和它对话，它可能会指引你并扩展，UI 会根据这个扩展和增长。所以仍然在扩展聊天和提示的基本想法，但以一种非常直观的方式教你，而不是这里一个工具提示，点击那里，尝试理解这个东西。

It's a really fun challenge and it's incredibly rewarding because like you said, there's no precedent, and you get to see your patterns become patterns that get copied too. But then you're also like, I don't know if that pattern was right, so you get to test to that scale through that method as well. I was chatting with Nate Parrot from the Browser Company the other day, and we were kind of debating back and forth. I was like, is the prompt box just the text field and a generate button really the best we can do? His opinion was actually that we're abstracting away a ton of complexity to just chatting, and that is actually maybe the design move. Maybe that is actually as simple as it needs to be. But then I think there's also a step forward here, which is there's a lot of stuff that you can do with these models. You can tweak all sorts of parameters, you can change the model type, and of course each model has either faster inference or can control the stability more and all these things. But you don't really want people to have to worry about a lot of those things. So even when I'm thinking about designing music, you can add lyrics, you can change the genre, you might be able to tweak just parts of the vibe of the track and the specific piece. But what is the most natural way to do that? Initially I could have started with just text and you describe every single aspect. But what that needs is that your user needs to know all those things to be able to craft a great prompt to get a good result. I was starting to think about this more in terms of a almost a dynamic island kind of prompt box, where if you need to change lyrics, it morphs into a prompt box that's basically best designed around how lyrics need to be edited and the parts of lyrics. If you're about to choose genre, instead of you thinking about it, it populates with pills of all the genres that you actually need to maybe want to play with or think about. And you could talk to it, and maybe it'll actually point you and grow, like the UI will expand and grow based on that. So still expanding on that base idea of chatting and prompting, but teaching you along the way in a way that's very intuitive, rather than tooltip here, click that, try to understand the thing.

我认为这就是用 AI 设计的关键：你需要仍然捕捉它的本质和魔力，而不把负担放在用户身上。我认为这是我们一直回到的问题。另一个我思考的辩论：很容易陷入使用经过长期测试的现有模式的陷阱，因为你认为那是可靠的。别误会，这有价值。例如，让我们谈谈搜索和搜索声音。如果你在搜索声音，经典设计模式：放一个搜索栏，有一些过滤标签，希望人们能找到正确的声音。但你实际进入那个页面在找什么？你有一个项目在脑海中。你可能在给一个孩子讲故事书，或者你在制作一部恐怖电影，想要一个反派的声音。所以这就是你真正需要描述的全部，我们应该在幕后找出你实际意思的标签，以及你应该应用的过滤器，还有我们认为最适合给你的口音。我认为这就是推动力：思考那些现有模式，但我们最初为什么发明它们？是因为我们有一些意图，我们将其提炼成旋钮和搜索字段等。而现在有了 LLM 和上下文理解，我们可以完全重写这些模式。

I think that's the thing with designing with AI: you need to still capture the essence and the magic of it without putting the burden on the user. I think that's the thing we keep going back to. Another debate I was thinking about: it's so easy to fall into the trap of using an existing pattern that's been tried and tested forever because you think that's reliable. And don't get me wrong, there's value to that. So for instance, let's talk about search and search for voices. If you're searching for a voice, classic design pattern: throw a search bar there, have some filter tags, and hopefully people will find the right voice. But what are you actually coming into that page looking for? You have a project in mind. You're probably telling a storybook for a kid, or you have a horror movie you're making and you want a voice for that horror movie for the villain. So that's all you should be really needing to describe, and we should figure out the tags that you actually mean behind the scenes, and the filters you should apply, and the accents we think are best to give you that. I think that's the push: think about those existing patterns, but why did we invent them in the first place? It was because we had some intent that we distill down into knobs and search fields and stuff. And actually now with LLMs and contextual understanding, we can completely rewrite those patterns.

我非常喜欢这个想法。几天前我和一家名为 Deson 的初创公司进行了类似的对话，讨论了类似的事情，比如拥有非常小的、基于提示上下文动态生成的对话，这对我来说开始变得更有趣。因为作为软件，你可以根据用户输入的初始内容直觉地判断他们想要实现什么，也许有一些更量身定制的可供性，你可以根据你所知道的生成来控制它。感觉这是一个动态界面的非常有趣的切入点，甚至在聊天 UI 之上构建。

I like that a lot. I was having a similar conversation a few days ago with a startup called Deson, and talking about similar things where having very small contained dialogues that are dynamically generated based on the context in a prompt, for instance, that starts to get a little bit more interesting to me. Because as a piece of software, you could intuit what a user is trying to achieve based on just the initial things that they're typing, and maybe there are some more tailored affordances that you can then generate to control that based on what you know. It feels like that's a really interesting entry point for a dynamic interface, building on top of the chat UI even.

是的，正如你所说，为什么我们最初最终制作了所有这些按钮、搜索字段和旋钮？那是我们对人们如何接近他们进入产品时的广泛目标的看法。

Yeah, and to your point, why did we end up making all these buttons and search fields and knobs in the beginning? That was our take on how people might approach the broad goal that they come into a product with.

就像，好吧，我想……我的意思是，这要追溯到图形用户界面的基础，对吧？我们以前在终端里打字，然后施乐和苹果决定你应该可以直接点击你想去的地方。同样地，为什么你不能直接说出你想做什么，然后界面做出反应，告诉你去哪里，并为你呈现出来？我认为这不只是纯语音，还需要一些触觉行为。但我们可以以前所未有的方式引导你，我觉得这非常非常令人兴奋。

It's like, okay, I want to... I mean, it goes back to the foundations of graphical user interfaces, right? We were typing things in terminal, and then Xerox and Apple decided you should just be able to click on the thing you want to go to. Similarly, why can't you just say what you want to do, and then the UI reacts and tells you where you need to go and serves it up to you? I think there's not just pure voice, but there's also a bit of that tactile behavior you still need. But we can guide you in a way we just couldn't do before, which I think is very, very exciting.

我们能深入探讨一下语音作为界面这个话题吗？我相信你和任何人一样都在思考这个问题。你在这个领域想象到的未来是什么样的？

Can we double-click on the voice as an interface thing? I'm sure you're thinking about this as much as anyone. What are the futures you're imagining in that world?

你知道吗，我觉得这很有趣，因为最近有很多很酷的产品问世，一开始炒作很热，然后要么就凉了，要么还在继续。我桌上还有一个，比如 Rabbit。我很喜欢它，觉得它是一件很酷的科技产品，但它还没有真正获得动力。我认为这是因为我们在社会规范方面还处于早期阶段。对很多人来说，在公共场合当着周围人的面跟一个设备说话，还是有点奇怪，对吧？这就是 Humane 也做了一些非常有趣的新颖体验和那个投影仪的原因，但同样，我还是觉得在人群中很不自在，不能直接说‘嘿，播放这个’或‘提醒我那个’，对吧？所以我认为一旦这成为更普遍的社会规范——而且我认为当苹果通过 AirPods 和更多主流设备开始采用这种界面时，这就会发生，我们最终会有更多 AR 眼镜，可以直接对它下达指令和交谈——语音将越来越成为我们习惯的界面。因为只需要看着前方，谈论你想做的事情，最重要的是，你周围的人也在做同样的事，就不再觉得奇怪了。那时我认为它才真正成为主流，并且需要一个像苹果这样的主流玩家来实现。但我记得电影《她》里的场景：一开始大家都在用设备滑动，然后某个时刻他们只是擦肩而过，但都在对着耳朵里的东西说话。一部 2013 年的电影能预测到这一点，我认为这完全会发生。我们就会拥有这样一个助手，它告诉我提醒事项，或者它看到了我见过的东西，并观察到这与我在日记里记下的内容有关，然后提醒我。是的，我认为这肯定会到来。前几天我和朋友开玩笑说，我未来想要的设计方式就是戴上某种可穿戴眼镜，然后我可以派我的智能体去制作一些软件和流程，然后在公园里通过可穿戴设备审查它们，给出一些指示，它们就继续工作。我退出专注状态，接住孩子扔给我的橄榄球之类的。最有效率的老爸！

You know, I think it's interesting because there are a bunch of cool products that have come out recently that had initial hype and then kind of tanked, or are still going. I still have one on my desk, like the Rabbit. I love it; I think it's such a cool piece of tech, but it didn't really gain momentum yet. I think that's because we're still early in the social norm side of it. It's still weird for a lot of people to just, out in the world, talk to a device when everyone's around them, right? And that's where Humane also did some really interesting novel things with the experience and that projector, but again, I still don't feel comfortable just being in a crowd and being like, 'Hey, play this' or 'Send a reminder for that,' right? So I think once that becomes more of a social norm—and I think it will happen when Apple leans into it more with AirPods and more of the devices that are mainstream start adopting that kind of interface, we'll have more AR glasses at some point that we'll be able to just say commands to and talk to—voice will become more and more the interface that we get used to. Because it's just about seeing what's ahead of you, talking about the thing you want to do, and everyone around you, most importantly, is also doing it, and it doesn't feel weird anymore. That's when I think it actually becomes the thing, and it needs a mainstream player, probably like an Apple, to make that happen. But I remember that scene from 'Her' where in the very beginning, everyone's kind of on these devices and scrolling, and then at some point they're kind of just walking by each other but they're talking to the thing in their ear. For a 2013 film to predict that, I think it's totally going to happen. We're just going to have this assistant that tells me my reminders, or maybe it's seen something I've seen and observed that relates to something I've put in my journal and reminds me about it. Yeah, I think that's coming for sure. I was joking with my friend the other day about how the way I want to be designing in the future is with some kind of wearable glass, and then I can just send my agents off to make some software and some flows, and then review them on my wearable while in the park, give some direction, and they just go off and keep working on it. I come out of that focus state and catch the football my kid throws at me or something like that. Most productive dad ever!

我喜欢这个。这仍然是《少数派报告》里的场景，对吧？我知道，我准备好了。说到这个，与一个研究团队合作是什么感觉？你几乎有两套输入：不仅仅是‘这是所有的用户问题和需求’，还有‘哇，我们刚刚发现了一个全新的东西——你想用它来实现什么？’那是什么样的体验？

I love that. It's still Minority Report, right? I know, I'm there. I'm ready for it. On that note, what's it like collaborating with a research team where you have almost two sets of inputs? It's not just 'here are all the user problems and needs,' but also 'whoa, we just unearthed this totally new thing—what do you want to make possible with it?' What's that like?

老实说，这就像你和一群魔术师一起工作，他们在我们的秘密实验室里想出了在描述音效的同时生成它的能力。就像，什么？所以我觉得这非常令人兴奋，因为回到之前的观点，很多事情你都没有模式可循，但你确实想解决问题。问题是：我花了无数时间浏览寻找适合我东西的音效，但我知道我想要什么。我带着我已经知道想要什么的意图去搜索。那么如果我能直接描述出来呢？这个了不起的研究团队，他们只是花时间推动模型进步。好吧，我们生成了语音。我们能生成任何形式的音频吗——音乐、音效？然后，如果能生成，我们能否清理并识别哪些音轨要移除，哪些要保留？即使是听写：如果我们能生成语音，我们能否也能理解语音，并创建一个更好的听写模型？所以他们会来找我们，说：‘嘿，我们有了这样的研究突破。我们应该如何部署它？’我喜欢他们这样处理问题——‘我们应该如何部署它？’——因为这给了设计师、工程师和产品人员真正思考人们会如何使用这个东西的机会。这个我们刚刚制造出来的极其技术性的模型——它最好的应用是什么？我认为我们经常开始的方式，有时甚至在内部，就是在设计之前先制作有趣的原型。就像，‘哦，我刚刚用了这个模型——当我们有对话式 AI 时——我做了一个入职机器人，可以告诉每个新员工 wiki 里的所有内容，甚至办公室的 Wi-Fi 密码。’然后你开始看到，‘哦，这是一个非常酷的应用。人们会如何自己制作一个呢？’然后你就可以开始设计练习来实现它。但这种黑客文化在 ElevenLabs 非常核心，现在几乎每个人都在玩 API，因为你可以用 Cursor 和其他工具。甚至通信和增长部门的人也在为自己制作很酷的原型。他们说：‘我刚刚做了个东西，可以跟画图聊天——还记得 MS Paint 吗？——现在你可以直接画画了。’我说：‘那是一个不可思议的原型。’这再次提醒我们，当人们拥有合适的工具时，创意可以来自任何地方。所以我认为这种黑客文化，加上看到人们正在制作的东西，让我们能够真正思考最好的设计方式，让任何人都能去制作那些东西。

Honestly, it's like you're working with a set of magicians who, in our secret lab, have come up with the ability to generate a sound effect while describing it. Like, what? So I think it's really exciting because, going back to previous points, you don't have patterns for a lot of these things, but you do want to solve the problem. The problem is: I spend an endless amount of time browsing for the right sound effect for my thing, but I know what I want. I'm searching with the intent of what I already know I want. So what if I could just describe it? This amazing research team, they just spend their time pushing the models forward. Okay, we've generated speech. Can we generate any form of audio—music, sound effects? And then, if you can generate, can you also clean and identify which stems to remove and which to take forward? Even with dictation: if we can generate speech, can we also understand speech and create an even better dictation model? So they'll come to us with, 'Hey, we've had this kind of research breakthrough. How should we go about deploying it?' And I love that they approach it that way—'How should we go about deploying it?'—because that gives the designers, the engineers, and the product folks to really think about how people would use this thing. This incredibly technical model that we've now made—what's the best application of it? I think often how we start, sometimes even internally, is by just making fun demos of stuff before we even design. So it's like, 'Oh, I just took this model—when we had conversational AI—and I made an onboarding bot that could tell every new employee everything in the wiki and even the Wi-Fi password of the office.' Then you start to see, 'Oh, that's a really cool application. How would people go about making one of their own?' And then you can start the design exercise of making it happen. But this hacking-away culture is very core to us at ElevenLabs, where almost everyone messes around with the APIs now because you can with Cursor and other tools. Even the comms and growth people are building amazing demos themselves for fun. They're like, 'I just made this thing where I'm chatting with Paint—remember MS Paint?—and now you can just draw stuff.' I was like, 'That is an unbelievable demo.' And again, a reminder that creative ideas can come from anywhere when people have the right tools. So I think that hacking culture, plus seeing what people are making, allows us to really think about the best way to design to enable anyone to go and make those things.

有没有一个你发布的项目或功能的例子，我们可以追溯从想法到最终上线的整个过程，以便了解你们是如何工作的，以及在 ElevenLabs 做设计师是什么感觉？

Is there an example of a project or a feature you released that we could trace the idea from inception all the way to shipping to production, just to get a sense of how you all work and what it's like being a designer at ElevenLabs?

是的，其中一个非常有趣的例子是我们在开发移动应用时，我们看到了 NotebookLM 的东西，但对我们来说，最合理的形态是移动设备。你有一个……

Yeah, one of the really fun ones was when we were working on the mobile app, and we had seen the NotebookLM stuff, but for us, the form factor that made the most sense was mobile. You have a...

很多阅读和内容都是在手机上随时随地进行的，我们想把它变成用我们声音的播客。我记得一开始我们基本上就是觉得，这是一种非常有趣的与内容互动的方式，而且我们的读者群增长得很好。我们看到很多人导入各种内容。我们该怎么做？移动团队的妙处在于，它确实是从这样一个问题开始的：我们如何给人们一种新的方式来与他们的内容互动，并且感觉非常无缝？你不应该去考虑选声音或做这个做那个。让我们真正把它变成这样，就像我们神奇地添加两个 AI 联合主持人，他们会把你有的任何话题变成一个有趣的故事，一个不同的叙事。然后团队就把任务交给设计，他们说，嘿，去尽情发挥创意吧，分享你觉得我们应该做的事情。

A lot of your reading and content on the phone on the go, and we wanted to turn that into podcasts with our voices. I remember the way it started was we basically were like, look, this is a really fun way to engage with content, and our readership is growing really well. We're seeing tons of people importing all sorts of stuff. How do we go about this? The beauty of the mobile team is it really starts with a problem like that: how do we go about giving people a new way to engage with their content, and it feeling incredibly seamless? You shouldn't have to think about picking voices or doing this. Let's really turn it into this like we're magically going to add two AI co-hosts that will take any topic you have and make it a fun story, a different narrative. And then the team kind of just hands off to design at that point, and they're like, hey, go and just kind of go crazy with ideas, share what you think we should do.

我记得我最初做了一个模型，我们的声音球慢慢发光，然后分成两个，表示现在你有两个联合主持人，他们互相绕着跳舞。我开始这么想的原因是，我知道我们生成的内容至少需要 30 到 40 秒，所以你甚至可以在无聊的加载状态中注入乐趣，你的联合主持人活过来，还有表演时间音乐在播放，让你准备好一个小型节目开始。我记得给团队看了那个，他们非常兴奋，然后我们开始想我们的联合主持人有什么酷名字。所以我们为不同的话题准备了不同的联合主持人，这显示了想法是如何不断增长，变成我们可以让这个功能活起来的各种创意方式。

I remember I initially made this mock of our voice orbs kind of slowly glowing and then splitting into two to show that now you have two co-hosts, and they're dancing around each other. The reason I started thinking about that was I knew that whatever we were generating would take at least 30 to 40 seconds to make, so you can even inject delight into a boring loading state where you have your co-host coming to life and it's got like showtime music playing, so you're getting ready for a little mini show to start. I remember showing that to the team, and they were so excited about it that we then started thinking about what are cool names for our co-hosts. So we have different co-hosts for different topics, and that kind of showed how the idea just kept growing into all the different creative ways we could make that feature come to life.

一旦我们满意了想要几个不同的联合主持人，因为有些话题需要严肃的，有些需要非常有趣的，我们就有了那个规格。我喜欢在很早的时候给工程师看早期的探索。他们会来回讨论并构建原型，这样我们就可以测试它的工作效果。尤其是对于这么早期的模型，这非常像一场实验游戏：你尝试，看看它在现实中的感觉，调整，然后来回修改设计。所以设计是非常动态的，一点也不静态。不是我做完交接就完了；而是，哦有趣，体验的那部分可能需要这里调整一下，那里调整一下，最终达到我们觉得好的最终产品。然后我们会在内部分享，看看人们如何生成各种播客之类的。模型在那时候也会出现各种有趣的 bug。我记得最好笑的一个是，我们的播客不知为何总是介绍 Sarah Chen 博士来评论事情。我们至今不知道 Sarah Chen 博士是谁，但现在每当有一个功能或我们不知道答案的硬技术问题时，我们就会说，我们去问 Sarah Chen 博士吧。这现在成了团队的一个梗。

Once we were happy with wanting a few different co-hosts because some topics need a serious one and some topics need a really fun one, we had that spec. I love showing the engineers very early on the early explorations. They'll go back and forth and build prototypes so we can test out how it's working. It's very much, especially with models that are this early, an experimentation game where you try it, see how it feels in real life, tune it, and go back and forth and change the design. So design is very dynamic and not static at all. It's not like I made this handoff; it's like, oh interesting, that part of the experience might need a tweak here, a tweak there, to eventually get to when we feel like we have a good final product. Then we'll share that internally and see how people generate all sorts of podcasts and things like that. All sorts of fun bugs show up with what the models do at that point too. I remember one of the funniest ones was our podcast for some reason kept introducing Dr. Sarah Chen to comment on things. We have no idea who Dr. Sarah Chen is to this day, but now whenever there's a feature or a hard technical problem that we don't have the answer to, we're like, let's go ask Dr. Sarah Chen. It's like the meme of the team now.

这真的是一个从问题开始的循环，设计去探索，非常实验性，非常像来回的原型设计。我在 Figma 中也做了大量原型，只是为了看看想法的感觉，让人们感受它在手机上的感觉并转化到桌面 UI。我甚至会使用我们的产品生成声音，然后录制那个原型并用声音编辑，这对原型来说听起来工作量很大，但我认为它让人们真正感受到它。它是实实在在的，因为他们会说，哦，所以对话听起来是这样的。一旦他们有了那个，我们就有了一个非常好的规格，知道需要做什么。我们实际上不写 PRD 或任何东西；这非常像一个来回的创意过程，直到我们对已有的构建感觉很好，然后我们就直接去做。

It's really the cycle of start with the problem, design goes and explores stuff, it's very experimental, it's very much like a back and forth prototyping. I prototype a ton in Figma too, just to see how ideas will feel and people get a feel for how it'll feel on the phone and translate to desktop UI. I'll even use our product to generate the voices and then record that prototype and edit it with the voices in, which sounds like a lot of work for prototypes, but I think it makes people really feel it. It's tangible because they're like, oh, so this is what a conversation could sound like. Once they have that, then we kind of have a very good spec for what we need to do. We don't actually write PRDs or anything; it's very much this back and forth creative process until we feel really good about the build that we have, and then we just kind of go with it.

是的，真的很酷。我的意思是，听你描述你的工作方式，感觉就像一种真正的实验文化。

Yeah, it's really cool. I mean, it just sounds like a real culture of experimentation, just listening to you describe the way that you work.

是的，非常如此。我们都是喜欢捣鼓的人，这很有趣。

Yeah, very much. We're all tinkerers, which is fun.

我的假设是，你们可能有一个非常广泛的用户群体。有些人可能像我一样，只是一个想捣鼓故事讲述的爸爸，而其他人可能是在专业地使用这个。所以你能谈谈你如何看待这项技术的正确抽象层次吗？

My assumption is that you probably have a pretty wide spectrum of users. Some people that are maybe like me, there's just like a dad that just wants to tinker with storytelling, where other people probably are using this professionally. So can you talk a little bit about how you think about the right level of abstraction for this technology?

这是个好问题。有时候更多的是你为每个用户角色创建界面。例如，我们相信应该始终有一个游乐场来摆弄模型，因为这能让你看到人们会以自由形式使用的所有方式。这就是为什么文本转语音仍然是一个非常开放的游乐场，音效也是一个开放的游乐场。当我们做音乐时，我们也会以同样的方式对待。但你可以在此基础上构建。我们知道有很多创作者和其他人只想生成一个声音，让它说些有趣的话，仅此而已。所以我们应该让这尽可能简单。但还有很多人有非常具体的声音和项目，他们真的在用心处理声音，我们知道我们需要给他们一个完整的编辑器，让他们可以分配多个说话者声音，调整精确的参数。这被作者、制作剧集等的企业公司和播客主使用。所以实际上就是：让游乐场向你展示与你的产品打交道的用户角色范围，然后与其中一些人交谈，问他们真正想要什么。你不想把所有这些都带进来，让你非常简单的游乐场变得复杂，因为那正是游乐场运作良好的原因。你实际上想把它拿出来，可能只是为那个用户角色为那个现有模型启动一个新界面。这就是我们应用的 Studio 版本的由来。这就是配音工作室的由来，它只专注于精确编辑那些配音，以确保翻译和配音的语言完全准确，因为现在有视频编辑团队只是在配音内容，他们只需要专注于那件事，其他什么都不用管。所以是的，就是把那个非常广泛的东西，然后为每个人制作非常小众的切片。

It's a great question. Sometimes it's more that you create interfaces for each of those personas. For instance, we believe that there always should be a playground to mess around with the models, because that allows you to see all the ways people are going to use it in a free-form way. That's why text to speech is still very much an open playground, sound effects is an open playground. When we do music, we'll very much treat it the same way. But then you can build on top of that. We knew there are a lot of creators and other folks who just want to generate a voice and have it say something fun, and that's it. So we should make that as easy as possible. But then there are a ton of people who have very specific voices and projects with voices that they're really working through, and we knew we needed to give them a full-fledged editor for that, where they could assign multiple speaker voices, tune the exact parameters. That's used by authors, by enterprise companies that are making episodes and stuff, and podcasters. So it's really like: let the playground show you the range of personas that are working with your product, then talk to some of them and ask them what are the things that they would really love. You don't want to bring all of that and complicate your very simple playground, because that is what made the playground work so well. You actually want to take that and potentially just spin up a new interface for that existing model for that persona. That's where our studio version of the app came through. That's where the dubbing studio came through, where it was just about precision editing of those dubs to make sure the language that was translated and dubbed was perfectly accurate, because you now have video editing teams that are just dubbing content and they just need to focus on that and nothing else. So yeah, it's about taking that very broad thing and then just making very niche slices for everyone.

嘿，我是 Red。经常有人问我最喜欢的产品，所以我花一分钟快速介绍一下我的技术栈。Deson 是我无需编码就能交付设计变更的方式。Framer 是我构建网站的方式。Genway……

Hey, it's Red. I'm constantly asked about my favorite product, so I'm going to take just one minute and give you a quick rundown of my stack. Deson is how I ship design changes without having to code. Framer is how I build my websites. Genway...

我想回到你之前提到的关于 Brex 的一些话题。你现在已经在 Plaid、Brex 和 ElevenLabs 这三家不同的超大规模初创公司工作过。基于这些经验，你认为在这样的环境中，作为一名设计师，需要具备哪些条件才能茁壮成长？

I want to return to some of the things that you were talking about around Brex. You've now been at Plaid, Brex, and ElevenLabs — three different hyperscale startups. Based on that experience, what do you think it takes to thrive as a designer in that type of environment?

每个环境都考验了我，也教会了我很多不同的东西，但可能有一条线索在所有环境中都最真实且贯穿始终：设计师不应该回避靠近业务。我认为真正理解客户是谁、我们为什么与他们合作、哪些合同最有潜力、我们如何定位公司——无论是面向小企业还是企业级客户等等——这非常重要，因为这能让你为自己所做的事情提出更有力的论据。我觉得有很多设计师会放弃，因为他们会说，‘哦，我提出了这个想法，但没成功，工程团队主导了现实。’但很多想法之所以被采纳，是因为它们有内在的商业价值。当然，设计还有另一个棘手的问题，那就是它相当主观，对吧？但我认为，通过建立量化方面和商业方面的可信度，你也能触及主观的一面，因为一旦有了信任，你就有更大的自由度去做你真正想做的事情，对吧？所以在 Plaid，公司非常鼓励你花时间与客户在一起，无论你担任什么角色都要去现场。作为一名设计师，我经常直接飞到法国和阿曼的客户那里，花一周时间了解他们的工作流程，并真正根据他们的需求进行设计，因为你成为了他们的专家，然后在构建过程中与他们形成了很好的反馈循环。在 Brex，你必须花时间与创始人和财务团队在一起，然后你意识到，你不仅仅是在做一个花哨的金融科技应用。现实是，创始人没有时间考虑财务问题——他们的公司卡或其他什么——所以实际上，体验需要尽可能对他们隐形。像在 Brex 中花费的时间这样的标准指标，你实际上并不想要。你真正想要的是他们能够快速完成工作，甚至根本不需要进入应用。同样，在 ElevenLabs，你有这些不可思议的模型和一群只想实现各种创意的人。暴露每一个旋钮并说‘每个人都应该做所有事情’是非常诱人的，但绝大多数人实际上只是有一个想法，他们想生成它，你应该让这变得尽可能简单。当然，你也会有高级用户，但你需要贴近业务和那些用户。设计师也不应该回避工程。还记得我们关于设计师是否应该写代码的永恒争论吗？这真是一个争论。现实是，对于设计师来说，每个人都很容易对你的工作发表评论，因为它是最容易接触的东西——每个人都可以看着它说，‘嘿，你应该调整那个按钮，你应该这样做，等等。’但为了拥有可信度和深入的理解，你实际上必须花时间与工程团队在一起，因为这能让你在他们那里建立可信度和尊重。然后他们会认为你理解他们正在经历的事情。业务方面也会明白你不是在空想。然后你就有自主权去真正执行。所以我想对设计师说，这份工作远不止是你的设计执行工作。我认为没有非技术性工作是低于你的。这是一份讲故事的工作——你必须把营销、品牌、工程、产品和战略之间的线索串联起来。我认为这是贯穿这些经历的一条线索。

Each environment tested me and taught me a lot of different things, but probably the one thread that was the most true in all of them and rang through throughout was: designers shouldn't shy away from getting close to the business. I think it's incredibly important to actually understand who the customers are, why we're working with them, which contracts have the most potential, how we want to position the company — whether it's small businesses or Enterprise or so on — because that just allows you to make stronger arguments for why you're doing the thing that you're doing. There are so many designers, I think, that give up because they're like, 'Oh, I proposed this idea, it didn't really work, and engineering overran what the reality is.' But a lot of these ideas get passed because they have some inherent business value. And of course, there's also the other tricky thing with design, which is that it's also pretty qualitative, right? But I think you get to the qualitative side through also building credibility on the quantitative side and the business side, because once there's trust, you then have more free rein to do the thing that you really wanted to do, right? So I think at Plaid, it was very much encouraged to spend time with the customers and go there no matter what role you had. Very often as a designer, I would actually just go and fly out to customers in France and Oman and spend like a week with them, just understanding their workflow and really designing for what they needed that way, because you become an expert for them and then you have such a great feedback loop with them as you're building things. At Brex, you have to spend time with founders and finance teams, and you realize, okay, you're not just making some flashy fintech app that needs to be flashy. The reality is founders have no time to think about finances — their corporate card or whatever — so actually the experience needs to be as invisible as possible for them. The standard metrics of time spent in Brex or whatever, you actually don't want that. You actually want them to be able to just get the job done really fast without being in there at all. Similarly with ElevenLabs, you have these incredible models and a whole set of people who just want to make all these creative things happen. It's very tempting to expose every single knob and say, 'Everyone should do everything,' but the vast majority of people actually just have an idea and they want to generate that, and you should make that as easy as possible. Of course, you'll have power users too, but you need to stay close to the business and close to those users. And designers should also not shy away from engineering. Remember that debate we've had forever about should designers code or not? It's such a debate. The reality is, for designers, it is so easy for everyone to comment on your work because it's the most approachable thing — it's the thing everyone can just look at and say, 'Hey, you should tweak that button, you should do this, whatever.' But for you to have that credibility and to have the deep understanding, you actually have to spend the time with engineering, because that allows you to build credibility and respect with them. Then they think you understand what they're going through as well. The business side understands that you're not just dreaming right there. And then you have the agency, I think, to really go ahead and execute. So I would say for designers, the job really is so much more than just your design execution job. I would view no non-technical job as beneath you. It's a storytelling job — you have to tie the threads together between marketing, brand, engineering, product, and strategy. And that, I think, has been a thread that's been true through those experiences.

那么，这些信念如何影响你对待招聘流程的方式，以及你身边人的类型，基本上是在一个全新的设计组织中？

How do those beliefs shape the way that you approach the hiring process then, and the types of people that you surround yourself with, basically given a blank slate of a design org?

我认为我非常喜欢那些直接展示而不只是说的人，你知道吗？我觉得我们已经习惯了作品集评审和面试流程，这些流程鼓励那种掩盖实质的戏剧性叙事。比如，我不想再看到一面墙的便利贴，然后你研究三个用户画像——我明白，那是工作，你知道的？那是工作，对吧？但我真正想看到的是，也许团队之间存在跨职能的鸿沟，而你帮助弥合了它，因为这才是公司日常运作中真正发生的事情，对吧？比如，你帮助两个团队看到并理解平台应该在哪些地方整合，或者他们需要在哪些地方做出妥协。或者你直接构建你的想法并展示出来，你不断进行实验。我喜欢那些喜欢动手和实验的人，因为那是我们的文化，我认为这对我们想要的东西也是净正面的。我们希望其他人也能融入这种创意组合。此外，我也喜欢那些背景多样、身兼多职的人，因为他们已经培养了对在这个阶段做许多必要事情的欣赏，你不会回避偏向行动。不要等待许可——你看到可以改进的东西就去做。所以我一直在寻找的设计师，是那些要么因为工作职责而不得不跨组织拓展的人。他们展示自己的想法，不介意分享，没有比这更好的了。他们对自己的创造力进行了大量实验，因为这个领域需要实验和创造力。有创始人背景的人也很棒，因为他们经历过解决困难问题的痛苦，知道真正从零开始构建需要什么。我想这就是初步的看法。

I think it's very much I love people who just show and don't tell, you know? I think we've gotten used to portfolio reviews and interview processes where we encourage the dramatic storytelling that hides the substance. Like, I do not want to see a wall of sticky notes again and then you study three personas — I get it, that's the job, you know? That's the job, right? But what I do want to see is how maybe there was a cross-functional gap between teams and you helped patch that through, because that's what's really going on in a company on a day-to-day basis, right? Like, you helped two teams see and understand where the platform maybe should have come together or where they needed to make a compromise. Or you just build your ideas and you show them, and you're just constantly experimenting. I love people who hack and experiment, because that's the culture we have, and I think that's net positive to what we also want. We want other people able to bring into that creative mix. And also, I love people who come from a variety of backgrounds, wearing many hats, because they've built an appreciation for doing the many things required at this kind of stage, where you don't shy away from biasing towards action. Don't look for permission — you just see something that you can improve and do it. So a lot of the designers that I've been looking at and looking for have been people who have either just had to stretch across their orgs because that's what the job commanded them to do. They show their ideas and they don't mind sharing it, and there's nothing like that. There's a lot of experimentation with their own creativity, because this is a field that requires that experimentation and creativity. And ex-founder backgrounds are also amazing because they've gone through the pain of figuring out the hard stuff, what it takes to really build something from the ground up. I think that's kind of the initial view.

自我意识也很重要，对吧？我一直觉得，我不想听到那种“AI 做得太多了”之类的回答。是啊，好吧，我们都这样。但我喜欢的是，当人们告诉我他们对设计的独到见解时。我听过有人跟我说：“其实我觉得我们不应该一开始就跟这么多用户交流，因为这会严重扭曲视角。”我就觉得，太好了，我喜欢这种人，他们非常相信自己的直觉，同时也对反馈持开放态度，对吧？这就是正确的要素：有直觉，但也能接受反馈，这样就能修正方向。太棒了。我们需要的是有品味的人，对吧？我们不想要那种完全依赖数据或者无法依赖数据的人，因为正如你所说，这是一个没有固定模式的模糊领域。你得自己摸索。

Self-awareness too, right? I think it's always like, again, I don't want the answer of "AI does too much" or whatever. It's like, yeah, okay, cool, like we all do. But I think it's like, I love it when people tell me their hot takes on design. I've heard, I've had people tell me like, actually, I don't think we should talk to this many users up front because it really skews the perspective. And I'm like, great, I love that this profile of a person actually believes in their intuition a lot more but is also open to feedback, right? And that's the right ingredient: intuition but has feedback so it will course correct. Amazing. Like we want taste makers, right? We don't want just people who are like can't rely on or only rely on data because like you said, this is an ambiguous space with no patterns. Like you have to figure it out.

那你们团队的工作方式呢？因为你们在流程、仪式以及协作方式上也是一张白纸。那么，有没有什么具体的东西是从你在 Palantir 或 Brex 的经历中借鉴来的，影响了你们设计这个组织的方式？

What about the way that you all work as a team? Because you're also given a blank slate in terms of processes and rituals and anything that collaboration looks like as well. So was there anything specific that you were pulling from your time either at Palantir or Brex that has influenced the way that you've designed the org itself?

好问题。Palantir 对我影响很大，它基本上很早就给了人们完全的自主权，让他们去做自己认为真正重要、能帮助团队的事情。我记得刚毕业的时候，入职才两周，我们在为一个情报机构工作，他们就说：“嘿，你能去给他们设计那个仪表盘吗？这个要呈报给该机构的负责人。”我当时想：“什么鬼？”你知道吗，这太疯狂了。但当你交付之后，那种自信心的提升，会让你之后带到工作中的能量都达到一个新高度，因为你感觉自己被完全赋予了权力。所以对于这个设计团队，我们的方法也是：你看到某个东西，一个需要讲述的故事，或者体验中需要改进的部分，直接去做就行了。没有设计系统警察，没有人会为了所谓的一致性而说“哦不，别那么做”。直接去尝试，我们随时可以调整。在这个阶段，一切都是可变的，没有哪个决定是不可逆的。所以我们非常开放地观察人们如何运作，然后拥有这样一个“杂牌军”团队。当然，我再次强调，等我们规模扩大到三四百人、需要更多流程的时候，这招可能就不管用了。但现阶段，我们的工程团队和设计团队都足够精简，我认为信任度很高。人们有好想法，就让他们去做。我们总能在最后修正和打磨。但你雇佣他们，就是因为你想赋予他们权力去做那件事。所以我认为这是 Palantir 方面很重要的一部分。

Yeah, great question. Palantir was a huge influence in that it had basically given people complete empowerment very early on to go and just do the thing that they thought was really important and would help a team. I remember as a new grad, very early on we were working for some intelligence agency and I was like two weeks in and they were like, "Hey, can you go and design that dashboard for them and it'll be presented to the director of this agency?" And I was like, "What the hell?" You know, it was insane. But the confidence boost after you deliver that just gives you a whole set of like, the energy you're going to bring to work is like next level after that because you feel fully empowered. And so with this design team as well, the approach is very much: you see something, a story we need to tell or part of the experience that needs to be better or whatever, just go do it. There's no design system police, there's no someone who's going to be like, "Oh no, don't do that" for consistency. Just go try the thing. We can always tweak it. Everything is changeable at this point and no decision feels irreversible at this stage. And so very much open to seeing how people operate and then having this team of like rag tag, especially again, I think this wouldn't work once we scale to a point of maybe 300-400 people where you need a bit more process. But we're at the stage where our engineering teams and our design teams are lean enough where I think it's high trust. People have great ideas, let them do their thing. We can always course correct and polish at the end. But you hired them because you wanted to empower them to do that thing. So that I think is a big part of the Palantir side of things which I think is really important.

Brex 方面我特别喜欢的是，那是我第一次完全体验远程公司，因为 2020 年疫情爆发，Brex 转向了全远程办公，他们做得非常出色，无论发生什么都能让所有人异步同步信息。所以，是的，你赋予人们权力，但也要让他们把决策反馈到一个中心位置。无论是通过一个快速的 Loom 视频展示他们的思考过程和做法，然后直接推进，还是与工程团队保持办公时间，因为会有很多请求你无法全职处理，你有一个决策文档，纸质记录是追踪事情的好方法，比如我们说过要做但没时间做的功能，后来我们又回来打磨了。或者任何新员工加入，我们只需给他们发一系列 Loom 视频让他们看，然后说：“这就是我们走到今天这一步的历程。”这对于入职培训来说是一种很好的方式，因为你只需要他们快速上手，给他们尽可能多的背景信息，这是让他们直观了解的最佳方式，而不是去翻 Figma 文件，然后想：“你做这个的时候到底在想什么？”你知道吗，这是一种非常有趣的视觉方式，可以帮助人们做到这一点。然后我想说，这带来的副产品是，人们自然而然地成为了讲故事的人，并且讲得越来越好，因为你不断在解释你的决策，向团队推销，在整个组织内推广。这是另一件我认为很重要的事：我希望每个人都能成为出色的讲故事者，因为这是非常高科技的东西，我们应该让它对每个人都易于理解。

The Brex side of things I really enjoyed was Brex was one of the first times I fully experienced a remote company because it was 2020 and the pandemic happened and Brex pivoted to being fully remote and they did an incredible job of keeping everyone up to date async no matter what was happening. And so yes, you have empowered people but always have them loop back with decisions being made in a central place. And so whether that is a quick Loom to show their thinking and how they approached a thing and then just running with it, or maintaining office hours with engineering teams because there are going to be many requests that you won't get to field on a full-time basis, and you have a decision doc of things that a paper trail has been an amazing way to track things like features we said we would do but we didn't have the time to do, and then we've come back and polished, or any new hire that joins, we just send them a series of Loom videos to watch and we're like, "Here's kind of the providence of how we ended up here." And that's been a great way to catch people up on onboarding because you just kind of need them to hit the ground running and give them as much context, and that's the best way for them to just see it rather than go through Figma files and wonder like, "What the hell went through your mind when you made that?" You know, it's just a very fun visual way to help people do that. And then I'd say the byproduct of that is people become natural storytellers and become better storytellers because you're constantly explaining your decisions and you're pitching that to teams and you're selling that across the org. And that is another big thing I think: I want everyone to be a great storyteller because this is really high-tech stuff that we should make approachable to everyone.

异步视频分享这个想法已经出现过好几次了，这仍然是一项每个人都在学习的技能。你有什么建议吗？你关注什么？或者甚至只是一点小建议或策略，听众下次为自己的团队做分享时可以应用的那种？

This idea of async video sharing has come up quite a few times and it's still kind of a skill that everyone's learning. Do you have any tips? What are you looking for? Or maybe even just a little tidbit of advice or a tactic that someone listening could apply the next time that they do a share out for their own team?

是的，我经常告诉人们的第一件事就是不要想太多。我觉得当你第一次在视频里看到自己，说错了一个词之类的，你会想：“哦，糟了，我要重新录一遍。”千万别这样。继续录下去。即使你不得不对自己说“第二条”，然后继续，这也没问题，因为你总是可以把那段剪掉。而且你最初可能说了一些话，如果重新开始，第二次可能就忘了。所以我认为完全不要想太多。这绝对是第一点。总是从“为什么”开始，从你要解决的问题开始。比如，你为什么走这条路？给人们提供背景信息，就好像他们在你的脑子里，是你的大脑一样。比如，你是怎么走到这一步的？因为这样你就向别人敞开了你的思维方式，接受其他视角，我认为这也很重要。如果你只给别人看结果，却不解释原因，那么我不知道你是怎么得出这个结论的，也不知道你是否做了正确的假设、正确的权衡，或者我可能看到的妥协，对吧？或者你看到了我没看到的东西，这给了我新的视角。所以我一直认为这很重要：分享你前期的思考。然后我总是喜欢用这样的框架来讲述故事：就好像我要向某人演示或推销它。我能让他们兴奋起来吗？所以即使对于播客这个例子，我会说：“酷，你有一篇你非常喜欢的文章，但如果你能对它有一个不同的视角呢？”而不是说：“你有了这篇文章，现在你可以用两个 AI 生成一个播客。”那不是“为什么”。而“为什么”是：这是你获得另一个视角的方式。我们喜欢这样吗？

Yeah, one of the first things I always tell people is don't overthink it. I think when you see yourself on video for the first time and you trip up on a word and stuff and you're like, "Oh crap, I'm going to re-record the whole thing," like don't. Actually keep going. Even if you have to say to yourself like, "Take two" and then continue, it's good because you can always cut that out. And you might have said something initially that you really would have forgotten about the second time if you had to restart. So I think don't overthink it at all. It's like definitely number one. Always start with the why, the problem you're solving. Like, why did you go down this path? Give people the context as if they were in your head and were your brain. Like, what did you kind of do to get you here? Because then you're opening yourself up to other perspectives on your way of thinking, which I think is also really important. If you just show people the thing but you don't explain the why, then I don't know how you got there and I don't know if you made the right assumptions or you made the right trade-offs or the compromise that maybe I'm seeing, right? Or you've seen something that I haven't and that's given me new perspective. So I always think that's like really important: share that upfront thinking. And then I always like to tell the story with the framing of as if I was about to demo it to someone or sell it to someone. Could I get them excited about it? And so even with the podcast one, for instance, I was like, "Cool, so you've got this article you really love, but what if you could have another perspective on it?" And so you've kind of, instead of saying "You got this article, now you can generate a podcast with two AIs," that's not the why. But the why is: here's how you get another perspective. And like, do we like that?

那条推文和那种框架方式，能让团队在公开谈论时的信息和方式上保持一致。所以试着把自己放在那个位置，就像你在推销这个东西：你会怎么推销它？因为这会迫使你看到它是否有价值。比如，我能为此感到兴奋吗？有没有真正会买它的受众？我认为这种框架很有帮助。

That thread and that way of framing this feature, and then that allows the team to also align behind the messaging and the way we would talk about it publicly. So try to put yourself in that like you're trying to sell the thing: how would you pitch it? Because that forces you to see if there's worth to this. Like, can I get excited about this? Is there an audience that would actually buy this? I think that framing really helps.

我喜欢看人们随机的实验。比如，给我看最终流程，然后给我看每一个分支的疯狂想法。很酷，对吧？我总告诉别人我不在乎你的 Figma 有多乱。即使在面试中，他们会说，“哦，我准备了演示文稿，但我的 Figma 有点……”我们就直接看 Figma 吧。我喜欢看你做了什么，而且不会因为乱而评判。我们都有自己疯狂科学家的思维方式。我只想看到工艺和思考。这就是我在乎的。

And I love seeing people's random experiments. Like, show me the final flow, and then just show me every branch off crazy idea. Cool, right? I always tell people I don't care how messy your Figma is. Even in interviews now, they're like, 'Oh, I prepared a presentation, but my Figma's kind of...' Let's just look at Figma. I love to see what you did, and there's no judgment on the messiness. We all have our way of mad scientist's way of thinking. I just want to see the craft and the thinking. That's all I care about.

好吧，在我放你走之前，对于那些可能对你 2025 年的疯狂科学家计划感兴趣的人，有没有什么未来的实验或你正在做的事情，或者你今年在考虑的事情，人们可以跟着看？

Well, before I let you go, for people who are maybe interested in your mad scientist plans for 2025, is there any future experiments or things that you're working on or things that you're thinking for this year that people can use to follow along?

是的，当然。Pedro 和我在 Twitter 上喜欢分享我们制作的 AI 应用和发布的所有东西。我们也总是分享为什么去做某个东西。比如，最近我做了一个 Perplexity 克隆版，但由 Gemini 搜索 API 驱动，因为我想看看 Google 是否还有实力，结果发现他们确实有。这是一个我为自己做的非常有趣的个人搜索应用。但我们可能会做的，可能不是要教一整门课——我觉得外面已经有很多好课了——而是为那些像我们一样喜欢捣鼓的人，他们只想要源代码，只想看到每一个未经筛选的提示词，每一个提示词，甚至包括我们对 Claude 大喊大叫的时候，以及我们如何走到那一步。我们可能会在那里开始一些东西，你可以关注我们，你会得到每一个提示词和所有源代码，这样你就可以把它变成自己的创意并重新组合。希望我们最终能形成一个小的黑客社区，他们也在里面分享，把想法放出来，互相激励。所以是的，我真的很期待能做得更多。

Yeah, no, for sure. Pedro and I on Twitter, we love sharing the AI apps that we make and all the stuff that we're putting out. We tend to also always share why we went and made something. For instance, recently I made a Perplexity clone but powered by Gemini search API because I wanted to see if Google still had what it took, and turns out they do. And it's a really fun kind of personal search app that I made for myself. But what we are going to do is probably we don't want to necessarily teach like a whole course—I think there are many great ones out there—but for the people who are tinkerers like us and just want the source code, just want to see every prompt literally unfiltered, every prompt, even the times where we shouted at Claude, and how we got there. We'll probably start something there where you can follow us and you'll have every prompt, all the source codes, so you can turn and remix it into your own ideas. And hopefully we end up making a small community of hackers who are also sharing within there, putting their ideas out there, inspiring each other. So yeah, I'm really excited about getting more.

是的，我也对此感到兴奋。与 AI 协作时的那种原始感非常重要，需要捕捉。所以即使只是听到你如何考虑拉开那层帷幕，也与我个人想学习的方式非常一致。

Yeah, I'm excited about that too. There's a rawness from collaborating with AI that is so important to capture. So even just hearing about how you're thinking about pulling back that curtain really aligns with how I personally want to learn as well.

不，我喜欢这一点。而且你说得对，关于原始感，很容易设计一门课程，其中每个提示词看起来都很完美，但过程中会有很多陷阱，你在与之斗争，人们可能就在那里放弃。我们想向人们展示，不，还有一点提示词的魔力，你可以克服它并做出很棒的东西。

No, I love that. And to your point, to the rawness, it's easy to craft a course where every prompt seems perfect, but there are so many pitfalls that come along the way where you're battling with it, and people can give up right there. And we want to show people like, no, there's a little bit more of prompting magic, and you can get past that and make something great.

甚至就像，我要为那些正在听、有点犹豫不决、好奇但可能还没动手做过什么的人强调一下，目标不是用 Claude 一次性搞定。它几乎从不那样工作。你总是需要打磨和调整。我的意思是，如果你一次性搞定某件事，也许我的提示词就是：再逼自己一点，做点更复杂的东西。

Even like, I'm just going to underline that for someone who is listening who's like kind of on the fence and is curious but maybe hasn't built something yet, like the goal is not to one-shot something with Claude. It almost never works that way. You always have to refine and massage. I mean, if you are one-shotting something, maybe my prompt would just be: push yourself a little bit more and do something more complex.

好了，Ammaar，这非常有趣。时间过得飞快。感谢你今天来分享，并让我们稍微了解了你正在思考的事情和你正在做的工作的幕后情况。

Well, Ammaar, this has been super fun. Absolutely flew by. Thanks for coming on today and sharing and giving us a little bit of behind the scenes of the things that you're thinking about and what you're working on.

非常荣幸。我关注你的播客和工作已经很久了，所以我很荣幸能来到这里。

Absolute pleasure. I've been following your podcast and work for so long, and so I'm honored to be here.