AI Interview Wiki
主题综述

AI 时代的品味与判断力 · Taste & Judgment

主题综述

更新日志

主流共识

第一点：AI 把"能力"商品化了——剩下的稀缺是品味、判断、自主性

这是贯穿全场的核心判断。Ivan Zhao 把它说得最锋利：

"Language model allows everybody to be a pretty good writer and programmer. So capability got normalized, democratized. And taste becomes still important. Like what's your value system? What do you want to bring to the world? Which direction do you want to go? That you cannot change as easily. And your will, how hard do you work? This you cannot change. So we optimize for the latter two today."
「语言模型使每个人都可以成为相当不错的作家和程序员。所以能力被正常化、民主化了。而品味依然重要。你的价值系统是什么？你想带给世界什么？你想朝哪个方向前进？这些你不能那么容易改变。还有你的意志，你有多努力工作？这一点你是无法改变的。所以我们今天优化这最后两个。」
Ivan Zhao (Notion) · Notion's Ivan Zhao: The Refounder

Fadell 给出的版本是一句行动指令——不要在认知上投降：

"Make great stuff, people, and don't think the AIs will. … We can use the machines, but don't cognitively surrender and make better stuff."
「做出伟大的东西吧，各位，别指望 AI 会替你做。……我们可以用机器，但不要在认知上投降——要去做更好的东西。」
Tony Fadell · Father of the iPod and iPhone on building taste, judgment

Bakaus 把 Fadell 这句口号落成了一个具体机制——"认知投降（cognitive surrender）"，并和无害的"认知委托（cognitive delegation）"划清界限：

"I'm using Google Maps and Google Maps tells me where I can go as quickly as possible. … But what if I let Google Maps decide where I want to go to? Now, I kind of like cognitively surrender to the application. And I think that's very true for LLMs if you're not careful. … You prompt something and then it creates a beautiful plan. And then it's eight pages long. You're not going to read through that plan, right? I mean, you're going to scroll through that plan and skim it. It's like, I guess the model knows what it's doing. … And then you just click OK. And so now you kind of like surrendered yourself to the process. … I think cognitive delegation is great. I think delegating to the model is great. But preserving that point of view, making sure that you're still the one driving is also really important."
「我使用谷歌地图，谷歌地图告诉我在哪里能够尽快到达。……但如果我让谷歌地图决定我想去哪里呢？现在，我有点像是对这个应用程序认知放弃。而我认为如果不小心，LLMs 也是如此。……你给出一个提示，然后它就会创建一个美丽的计划。然后它有八页长。你不会逐页阅读这个计划，对吗？我的意思是，你会快速浏览那个计划。像是，我想模型知道自己在做什么。……然后你只是点击确定。所以现在你有点像是将自己的主权交给了这个过程。……我认为认知委托是很好的。我认为委托给模型也是很好的。但保留那个观点，确保你仍然是驾驶者，这也非常重要。」
Paul Bakaus · What Happens to Design After AI

Mark Chen 从前沿研究实验室的内部给出了同一判断的另一种形态——即便模型已经能端到端做研究，稀缺的仍然是"品味"这一层，而它恰恰最难教给模型：

"I think both at OpenAI and at other labs, you're starting to see a lot of the work become mostly orchestration focused, right? Like the researchers coming up with ideas. And the model's great enough to do the implementation execution by itself. … And that's why you still need the researchers coming up with the ideas. … It's going to be hard to teach the models good taste. We noticed that."
「我认为在 OpenAI 和其他实验室，你开始看到很多工作主要集中在编排上，对吧？就是研究人员提出想法。而且模型足够优秀，能够自己完成实施执行。……这就是为什么你仍然需要研究人员提出想法。……教会模型良好的品味将会很困难。我们注意到了这一点。」
Mark Chen (OpenAI) · Cooking with OpenAI's Research Chief

第二点：可测量的东西不是全部——警惕"数据驱动"在创新上的陷阱

Fadell 和 Ive 从两端撞到同一个判断。Fadell 讲 1.0 产品：

"If you try to do data-driven decisions all the way along, you're either not doing a differentiated product because you're taking data from another thing or you're just getting bullshit data."
「如果你一路都试图做数据驱动决策，那你要么没在做差异化产品——因为你的数据是从别的东西上拿来的——要么你拿到的只是垃圾数据。」
Tony Fadell · Father of the iPod and iPhone

Ive 把它上升到一个更尖锐的指控——可测量性是一个谎言：

"We spend all our time talking about attributes because we can easily measure them. Therefore, this is all that matters. And that's a lie."
「我们把所有时间花在谈论属性上，因为它们容易测量。于是'这就是全部重要的东西'——而这是个谎言。」
Jony Ive · A conversation with Jony Ive

第三点：AI 内容泛滥反而让"真实性"升值

Reddit 的 Huffman 给了这条最直接的市场侧证据：

"People can feel authenticity."
「人们能感受到真实性。」
Steve Huffman (Reddit) · How Reddit Went From $12M to $2.2B

Maeda 从设计一侧独立撞到了同一结论——他把它命名为"人味（human smell）"，并直接点出溢价的来源是"人类的信任与担责"：

"The reason why someone is willing to pay that much more for that bespoke better thing is they're paying for human trust and accountability. … That human ability, that human smell could become even more valuable than ever before."
「一个人愿意为那个定制的、更好的东西多付这么多，是因为他在为人类的信任与担责付费。……那种人的能力、那股'人味'，可能比以往任何时候都更值钱。」
John Maeda · What Happens to Design After AI

这就把 Huffman"人能感知真实性"从社交媒体的市场直觉，升级成了一条跨领域（社交内容 / 软件设计）的共识——AI 内容越泛滥，可辨识的"人的意图与担责"越升值。

分歧在哪

大家都同意"品味/判断是 AI 时代的稀缺资源"，但三个问题上分歧明显：品味是天生还是习得、AI 时代该收紧还是放开控制、品味体现在哪一层。

分歧一 · 品味是天生的，还是练出来的？

Ivan Zhao 把品味和自主性当作不可改变的东西——所以他的招聘是哑铃型（资深架构师提供品味，超级初级提供自主性，砍掉中间）：

"Hire for super junior, super senior. … Barbell shape, right? Because the senior provides taste. … A good engineer managing four to six coding agent[s] at once, right? But a good, really senior architectural engineer can manag[e] two to three junior interns or engineer[s], [and] they each manage two or three."
「招超级初级和超级资深。……哑铃型，对吧？因为资深的人提供品味。……一个好工程师同时管四到六个 coding agent，对吧？但一个好的、真正资深的架构工程师能管两三个初级实习生/工程师，这些人每个再各管两三个。」
Ivan Zhao (Notion) · Notion's Ivan Zhao: The Refounder

Fadell 的"informed gut"恰恰相反——gut 前面那个 *informed* 是练出来的，是经验的产物，不是天赋：

"When you're doing a 1.0 of anything, if you're doing anything that matters, and it's a 1.0 and it's a new category or it's a new device the world hasn't seen before, you have very few analogs that you can use to make data-driven decisions. … You have to have one or two or a very small set of people who are charged with making the opinion-based decisions."
「当你在做任何东西的 1.0——如果你做的事情很重要——而它是 1.0、是一个世界前所未见的新品类或新设备时，你几乎没有可用来做数据驱动决策的类比。……你必须有一两个、或极小的一组人，专门负责做这些基于判断的决策。」
Tony Fadell · Father of the iPod and iPhone

Ive 又给了第三种——品味是一种辨别力，而它最大的敌人是急于表达观点的人：

"What kills most ideas, I think, [is] people desperate to express an opinion. And it's really, let's be very clear, opinions aren't ideas."
「我认为，杀死大多数 idea 的，是那些急于表达观点的人。说清楚一点——观点不是 idea。」
Jony Ive · A conversation with Jony Ive

Maeda 给了第四种、也是最锋利的——品味不是个人属性，而是文化-历史的沉淀物，来自材料的稀缺：

"Taste is always cultural and different cultures have quote-unquote higher taste, specifically because they've usually been around longer. … When raw material is scarce, we tend to make it more precious and design things well. Taste emerges through scarcity. … I think what's interesting about this era is that this idea of taste doesn't fit when all the materials available to everyone."
「品味永远是文化性的，不同文化有所谓'更高的品味'，恰恰因为它们通常存在得更久。……当原材料稀缺时，我们倾向于让它更珍贵、把东西设计好。品味源于稀缺。……这个时代有意思的地方在于，当所有材料对每个人都唾手可得时，这套'品味'的观念就不成立了。」
John Maeda · What Happens to Design After AI

这一条其实是"习得派"的一次升级：品味不是天赋（反 Ivan），但也不是个人靠经验就能练出的 informed gut（超出 Fadell）——它是几百年材料稀缺压出来的集体成熟度。而 AI 把材料变得无限丰裕，恰恰抽掉了旧品味赖以形成的稀缺土壤。

Mark Chen 则第一次给了"品味可教"一个可操作的实证——他把研究品味当作能被复现动作练出来的技能，而不是玄学：

"The best mechanism I've found for developing that [research taste] is really just papers that you really look up to, just try to fully replicate it. … I learned so much just trying to replicate the training curves exactly, get to the exact amount of training loss or perplexity that the paper is hinted towards. It teaches you a lot of techniques that people don't really talk about."
「我找到的培养它（研究品味）的最好机制，真的就是拿你非常仰慕的论文、试着完整复现。……我光是尽力去精确复现训练曲线、逼近论文暗示的那个训练损失或困惑度，就学到太多东西。它教给你很多没人明说的技巧。」
Mark Chen (OpenAI) · Cooking with OpenAI's Research Chief

五个人对"品味从哪来"给了不兼容的答案：天生的（Ivan）、练出来的（Fadell）、辨别出来的（Ive）、文化稀缺沉淀的（Maeda）、靠"精确复制大师"练成的（Mark Chen）。值得注意的是，Mark Chen 那条恰好接上了我原来在"还想知道什么"里追问的"品味可教的真实案例"——复现经典论文就是那个可观测的训练路径。

分歧二 · AI 时代该收紧控制，还是放开？

Fadell 主张在关键细节上收紧——战略性微管理 + 反对全 AI 生成代码：

"You have to have, for lack of a better word, tastemakers. … We are the person or the team who is going to make those opinion-based decisions. Of course, some people aren't going to like it and say, 'I'm sorry, this is a benevolent dictatorship. This is what's going to happen and this is the vision.'"
「你必须有——找不到更好的词——'品味制定者（tastemakers）'。……我们就是那个要做这些判断型决策的人或团队。当然有人会不爽，那就说：'抱歉，这是一个仁慈的独裁。事情就这么定，这就是愿景。'」
Tony Fadell · Father of the iPod and iPhone
"Anybody who looked at the code, who's a real software architect and engineer, threw up … They're like this stuff is brittle. … This should be layered in four or five, actually 12 or 15 different sub-functions. … There is this dichotomy of fast and throwaway. It's called fast fashion. We got fast software. But software, if you're going to build a real company, can't be throwaway."
「任何看过那段代码的、真正的软件架构师 / 工程师都吐了……他们说：这玩意儿太脆了。……这本该拆成四五个——其实是 12、15 个——不同的子函数。……这里有一个'快且一次性'的二分——就像快时尚。我们现在有了'快软件'。但软件，如果你真要建一家公司，不能是一次性的。」
Tony Fadell · Father of the iPod and iPhone

Ivan Zhao 恰恰相反——他拥抱失控，把和 AI 一起开发比作酿啤酒：

"Building classic software is like engineering bridge. If you can't design it, you usually can't build it. Fairly predictable. … Building with language model back then and somewhat still is, it's like brewing beer. You can't really predict the things."
「构建经典软件就像造桥——如果你设计不出来，通常也就造不出来，相当可预测。……而用语言模型构建，当时是、现在多少也还是，像酿啤酒——你没法真正预测那些东西。」
Ivan Zhao (Notion) · Notion's Ivan Zhao: The Refounder

Fadell 要在重要的地方握紧（造桥的精确），Ivan Zhao 要在过程里放开（酿酒的不可控）——两人都在谈"怎么和 AI 一起做产品"，姿态正好相反。这跟 founder-mode 主题里 Chesky（纵深握紧）vs Foroughi（放开信任）的对立是同构的。

分歧三 · 品味体现在哪一层？

Ive 把它放在道德/精神层面——做东西是在表达"你是谁"、关不关心用户：

"When somebody unwrapped that box and took out that cable, and they thought somebody gave a shit about me, I think that's a spiritual thing."
「当有人拆开那个盒子、拿出那根线，心里想'有人在乎我'——我觉得那是一件精神层面的事。」
Jony Ive · A conversation with Jony Ive

Cannon-Brookes 把它放在商业差异化层面——设计是难以复制的护城河：

"When you have any major technological transition, we're back to an era of fundamental design. … In the AI era, in an era where arguably software gets cheaper to create, we can argue about that maybe, but there will certainly be far more software. … [We're] really doing a lot of foundational design around delivering AI to end-users and customers that don't have to understand what the words probabilistic and deterministic mean."
「每逢重大技术变革，我们就回到一个'基础设计'的时代。……在 AI 时代——一个可以说软件造起来更便宜的时代，这一点或许还有得争——但软件一定会多得多。……我们在做大量基础设计——把 AI 交付给终端用户和客户，让他们根本不必懂'概率性''确定性'这些词是什么意思。」
Mike Cannon-Brookes (Atlassian) · 20VC: Atlassian CEO

Huffman 把它放在市场可感知层面——真实性是用户能"闻到"的、会用脚投票的东西。

同一种"品味/判断"，Ive 当作伦理、Cannon-Brookes 当作护城河、Huffman 当作市场信号——三种价值归属，没人调和。

分歧四 · AI 是抬高地板，还是拉大差距？

新增的三篇设计 / 创作访谈里，所有人都同意一句话"AI raises the floor（抬高地板）"——但对"抬高地板之后"的判断分成了两派。

Bakaus / Maeda 是"抬地板→抬天花板"的乐观派：地板抬高把机械劳动清走，正好把人解放去做那最后 10–20% 的独特工艺，他们赌这会开出一个新的"手工艺时代"：

"We're kind of like raising the floor in areas that can be mechanical. … humans should be able to raise the ceiling of what they can do."
「我们是在把那些可机械化的领域的地板抬高。……而人类应该能去抬高自己能做到的天花板。」
Paul Bakaus / John Maeda · What Happens to Design After AI

但 Maeda 自己在同一场对话里给这份乐观加了一个冷酷的市场注脚——能抬天花板的人本来就极少，而且大众市场不为它买单：

"The number of people who will raise the ceiling will not be a high number. And it won't be that valuable in the mass market. So I think it'll never be automated. … Only problem, the customer base will be smaller. … do we know letterpress, letterpress printing? Oh my gosh, I love it. But not many people will pay for it."
「能抬天花板的人不会有很多。而且它在大众市场上不会那么值钱。所以我认为它永远不会被自动化。……唯一的问题是，客户群会更小。……我们还认得活版印刷吗？天哪我爱它。但没多少人会为它付钱。」
John Maeda · What Happens to Design After AI

Photolabs 的 Xia 说得更直白——地板抬高的同时，顶尖和平均之间的差距反而拉大：

"With AI tools, they're definitely raising the lower bar just so that, you know, even someone like me can create something pretty, pretty good with the AI tools. But I also think that the gap between the very best artists and average artists is going to be even bigger."
「有了 AI 工具，它们无疑提升了最低门槛，以至于像我这样的人也能用 AI 工具创造出相当好的东西。但我也认为，优秀艺术家与普通艺术家之间的差距会变得更大。」
Zach Xia (Photolabs) · Building AI for Creators

Luma 的 Tancik 紧接着补了一个更具体的注脚——产出"垃圾"和产出惊艳的是同一套工具，差别只在用工具的人：

"You've seen AI generations that are bad, right? I think we've all seen them that are bad. I can guarantee the same tools that made those bad generations, they've also made generations that you would think are amazing."
「你见过糟糕的 AI 生成的内容，对吧？我想我们都见过一些糟糕的内容。我可以保证，造成那些糟糕生成的同样工具，也创造出你认为非常惊艳的作品。」
Matt Tancik (Luma) · Building AI for Creators

而他们对"AI 时代创造力是什么"的回答，把品味从"审美"重新定义成了"导演能力"——不是掌握工具，而是指挥工具的那个方向性头脑：

"Creativity is building a story. The tools alone aren't a story. Someone has to direct them. … It's not about mastering those tools. It's about directing an agent who can use those tools to achieve your creativity."
「创造力是构建一个故事。工具本身不是故事，得有人来指挥它们。……重点不在掌握那些工具，而在指挥一个能用这些工具去实现你创意的代理。」
Matt Tancik / Zach Xia · Building AI for Creators

这一节和"我的看法"里那条主论断（品味会让优势向已有品味者集中）直接互证：Xia 的"差距拉大"是一线产品经验里跑出来的观测（Tancik 从工具侧补证），而不再只是我从 Fadell 外推的推断。乐观派（Bakaus）和集中派（Xia/Tancik）没有被调和——同一个"抬高地板"，一派读成普惠，一派读成分化。

暗流 · Ivan Zhao 的"不懂历史就不懂人性"

Ivan Zhao 给了一句跟所有人都不冲突、但所有人都没说的话——技术圈最大的盲点是不懂自己的历史：

"… tech is largely a technology and tinker culture. And tinker culture doesn't even know the history. … So tech is like industry doesn't know its past. If you don't know its past, you don't know history, which is humanity."
「……科技大致上是一种技术和修补文化。而修补文化甚至不知道历史。……所以科技界就像个行业，不知道自己的过去。如果你不知道自己的过去，你就不知道历史，而历史就是人性。」
Ivan Zhao (Notion) · Notion's Ivan Zhao: The Refounder

这句话给整个"品味"主题埋了一条更深的线——所谓品味，可能本质上是对人性和历史的理解深度，而不是审美天赋。

都没说透的

我的看法

判断（不是事实）：这个题目里"品味是 AI 时代的护城河"是真的，但它正在被讲成一种安慰剂——一群本身拥有顶级品味的人，互相确认品味重要。真正有价值、也最被回避的问题是分配：如果品味像 Ivan Zhao 说的不可改变，那"AI 让能力民主化"的另一面是品味的稀缺性反而被放大——能力不再是门槛后，品味成了唯一的门槛，而它分布得比能力更不均。我赌 Fadell 那条"informed gut——可练的判断"比 Ivan 的"天生不可改变"更接近真相，但练成它需要的时间和高质量反馈，恰恰是 AI 加速的世界最稀缺的。所以实际结果可能是：品味确实是护城河，但它会让优势进一步向已经拥有品味的少数人和组织集中——这跟"民主化"的叙事正好相反。

把握程度：中等偏上（较上一版上调）。最强支撑是"可测量不是全部"+"真实性升值（Huffman 的市场信号 + Maeda 的'人味/担责'从设计侧独立佐证）"这两条跨多个独立顶级从业者的共识；此外，我原来最弱的那条"品味会导致优势集中"，现在被 Xia 的一线产品观测直接顶上了——"AI 抬高地板的同时拉大顶尖与平均的差距"不再只是我从 Fadell 外推的推断，而是做创作工具的人跑出来的经验。仍最弱的一环是"练成品味需要的高质量反馈恰恰在 AI 加速的世界里最稀缺"——Mark Chen 的"复现经典论文"给了半个反例（研究域里反馈是冷硬可得的），说明在有 ground truth 的域里这条链未必成立；它是否只在无真理的品味域（设计、创意）里才锁死，语料还没接完。

还想知道什么

取材

不切题、未计入综述：Uncapped #53 (Trae/Delian, Founders Fund) 390ea6160e7181159c6ac4cc0b0c7a99、Benchmark's AI Bets 390ea6160e7181848860ca758a56d9ab——两篇的"conviction/taste"是 VC 投资语境与"software tastes like chicken"财报双关，不属于本主题的品味/判断力。

继续读
线索
阅读线索：agent 不到 9 nines 可靠，那应该卖给企业还是消费者
线索
阅读线索：如果 coding 是 commodity，AI startup 的护城河长在哪儿
线索
阅读线索：推理成本一年跌 10×，AI app 该按什么定价
线索
摘报
线索
摘报
线索
摘报