Kevin Kelly

每周链接，2025年11月17日 || Weekly Links, 11/17/2025

2025-11-14T22:37:00+00:00

这不是你每天都能看到的标题……虎鲸想出了一种聪明的策略来围攻大白鲨并吃掉它们的肝脏，这一行为首次被视频记录下来。
我出现在这个关于HyperCard的新闻报道中，那时我还很年轻。HyperCard输给了万维网，但在许多方面都优于后者。

---------------

This is not a headline you see everday…Orcas have devised a clever trick to gang up on great white sharks and eat their livers, which has been captured on video for the first time.
I appear in this news report on HyperCard as a young man, long ago. HyperCard lost out to the WWW but it was superior in many ways.

中国自由行必备指南 || Essentials for Independent Travel in China

2025-10-29T18:40:06+00:00

你应该去中国。中国地域辽阔，非常多样，安全，出行方便，价格实惠，有趣，与你想象的不一样，而且在世界上的地位越来越重要。亲自去看看吧。然而，与中国其他地方不同，中国有自己的平行应用程序集，你必须在那裡使用它们。以下是我认为在中国独立旅行必备的移动应用程序。一个好习惯是出发前在中国以外下载这些应用程序，因为大多数应用程序都受到“防火长城”的限制。确保你下载的是应用程序的“国际”版本（如果有的话），这样它就可以使用英文。（中国现在有一个新的免签10天选项，使访问变得更加容易。有关这种过境签证的详细信息，请参见这篇博客帖子。） Airalo 现代的中国生活离不开手机。你需要一个可靠的本地网络连接来使用这些应用程序。一些美国的移动套餐，如T-Mobile或Google Fi，会自动为你提供在中国的手机信号，而像Verizon这样的套餐则每天收费高昂。对于其他人，或者为了避免额外费用，你需要一张本地SIM卡。你可以在抵达机场时购买SIM卡，但最近的手机型号支持eSIM，这是一种你只需下载的应用程序，而不是实体卡。你为不同的数据套餐付费。有许多eSIM初创公司，如Saily、Holify和Airalo。你也可以从Trip（见下文）购买eSIM。目前，为所有这些eSIM加载数据都比应有的更繁琐，因此我建议你在抵达中国之前就下载好eSIM。一些eSIM的好处是它们通过香港或新加坡路由流量，因此它们可以充当内置的VPN，这意味着你不需要在手机上安装其他VPN应用（见下文）。我在亚洲和欧洲国家一直使用Airalo作为eSIM，它在中国也能正常工作。重要的是，Airalo内置了VPN，因此在中国，我可以轻松地在手机上使用电子邮件、聊天AI和新闻网站。 Express VPN 防火长城会阻止Gmail、ChatGPT、社交媒体、YouTube和主要新闻网站等。为了绕过你手机上这个令人沮丧的限制，使用一个好的eSIM。为了在你的笔记本电脑上绕过这个限制，你需要一个VPN。VPN在与防火长城的猫鼠游戏中不断更新，但它们并不总是有效。我在使用Express VPN的付费版本（每月8美元）时在中国取得了成功。其他旅行者和许多在中国生活的外籍人士更喜欢AstrillVPN，虽然价格更高，每月15美元。确保在出发前更新到最新版本，因为它们经常更新。最好下载另一个版本作为备份，以防你首选的版本停止工作，这种情况并不罕见。 Alipay 中国已经实现了无现金支付；现在没有人再使用现金，除了那些绝望的游客。而且大多数商店都不接受信用卡。中国有两种广泛使用的数字现金系统，即微信支付和支付宝。我推荐使用支付宝，因为它们现在让外国人很容易地设置与信用卡绑定的账户。支付宝应用会生成一个二维码，商家可以扫描，或者你可以扫描他们的二维码。支付宝是一个超级应用，里面包含许多子应用，如网约车、共享单车、门票和翻译。这个应用是必不可少的。 Didi 中国没有Uber或Lyft，但有Didi，其功能类似。你可以安装Didi应用，但更好的是，Didi已经整合在支付宝应用中。因此你不需要设置任何新东西。在支付宝中查找Didi即可。非常方便。我更喜欢使用Didi而不是出租车，以避免在沟通目的地时遇到困难。常见的做法是当你进入出租车时，报出你手机号码的最后四位数字来确认你的行程。 WeChat 使用微信来与中国人进行短信、通话和聊天，无论他们处于什么层次。这是中国每个人都有的一款应用。当你分别或见面之前，每个人都会要求你的微信账号。它无疑是与中国人沟通的首选方式。你们可以通过扫描彼此的二维码来交换联系方式。在中国以外安装微信可能有些麻烦，但这是个好主意。 Amap Amap是中国最好的地图应用。（Google和Apple地图在中国的更新频率和详细程度不如Amap。）除了导航功能，Amap还可以用来搜索附近餐厅或景点的英文信息。它有点像Yelp的简化版。它还会提供城市公共交通和地铁的详细路线和连接信息，这在实际中非常有用。（使用支付宝购买车票。）你需要在手机上离线下载该应用，或者使用VPN来获取英文版本。 Apple Translate 你可以使用Google翻译应用（搭配VPN），但通常我会使用iPhone内置的Apple Translate应用。（支付宝也有一个翻译应用。）你至少需要其中一个。你可以将手机或相机对准标志或菜单上的文字进行翻译，或者使用“对话”模式将语音转换为文字，甚至进行语音对话的翻译。 Trip 在中国预订航班和高铁时，使用Trip。它们可靠，几乎涵盖所有选项。高铁几乎可以到达任何地方，但国内航班便宜且众多。你可以通过Trip预订两者。此外，Trip在中国预订酒店也最方便。出于习惯，我仍然使用Booking.com，它也适用于酒店（不适用于高铁或航班）。Trip在非旅游地区有最好的选择，而Booking.com则适用于大多数城市。
---------------

You should visit China. It’s vast, very diverse, safe, easy to get around, inexpensive, interesting, not what you expect, and increasingly important in the world. Go see for yourself.

However, unlike the rest of the world, China uses its own parallel set of apps that you will need to operate there. Here are what I consider the essential mobile apps for independent travel in China. A good rule of thumb is to download your apps outside of China before you leave, because most are behind their great firewall. Make sure you are downloading the “international” version of the app (if it has one) so that it uses English.

(China has a new visa-free 10-day option that makes it easier than ever to visit. For details of this transit visa see this blog thread.)

Airalo

Modern China lives on the phone. You need a good solid local connection for your apps. Some US-based mobile plans like T-Mobile or Google Fi will automatically give you cell coverage in China, and others like Verizon will charge you a premium $12/day. For everyone else, or to avoid the surcharge, you’ll need a local SIM. You can get a SIM card at the landing airport but recent phone models permit eSIMS, which is a mere app you load instead of a card. You pay for different data plans.

There are a bunch of eSIM startups, like Saily, Holify, and Airalo. You can also purchase an eSIM from Trip (see below). Loading an eSIM at the moment is way more cumbersome than it should be for all of them, so I’d recommend you download your eSIM before you arrive. One of the benefits of some of the eSIMS is that they route their traffic through Hong Kong or Singapore so they act as a built-in VPN, which means you can avoid needing that app on your phone (see below). I’ve been using Airalo as an eSIM in Asian and European countries, and it works fine in China. Importantly, Airalo has a VPN built in, so in China I could do my email, chat AI, and newssites on my phone effortlessly.

Express VPN

The Great Firewall will block Gmail, ChatGPT, social media, YouTube, and major newssites among others. To get around this frustrating block on your phone, use a good eSIM. To get around it on your laptop you’ll need a VPN. VPNs are in a cat-and-mouse race to keep the channel open, and they don’t always work. I’ve had success in China using a paid version of Express VPN ($8 per month). Other travelers and many expats living in China prefer AstrillVPN, which is more expensive at $15 per month. Make sure you have updated it before you leave with the latest version, which is updated frequently. It’s a good idea to download a second version as a backup in case your main choice stops working, which is not uncommon.

Alipay

China is cashless; no one uses cash anymore, except desperate tourists. And most stores do not accept credit cards. There are two widespread digital cash systems, WeChat and Alipay. I recommend using Alipay because they now make it very easy for foreigners to set up an account linked to your credit card. The Alipay app will generate a QR code the seller will scan, or you can scan their QR code. Alipay is a super-app with many sub apps inside it, such as rideshares, bikeshares, tickets, and translation. This app is an essential must-have.

Didi

There is no Uber or Lyft, but there is Didi, which works the same. You can install the Didi app, but even better, Didi is available within the Alipay app. So you don’t need to set up anything new. Look for Didi inside Alipay. Very convenient. I prefer using Didi to a taxi to eliminate the challenge of communicating my destination. The common protocol is to recite the last four digits of your phone number when you enter the car to confirm your ride.

WeChat

Use Wechat to text, call, chat with other Chinese at all levels. It’s the one app everyone in China has. Everyone will ask you for your Wechat account when you part or before you meet. It by far is the preferred way to communicate. You scan each other’s QR codes to exchange numbers. Installing WeChat outside of China is cumbersome, but a good idea.

Amap

Amap is the best map app in China. (Google and Apple Maps are not as current, nor as detailed as Amap.) Besides navigation purposes, Amap is also useful for searching for what restaurants or sights are nearby in English. It kind of works like Yelp-lite. It also gives you detailed instructions and connections for public transit and subways in a city, which is a real lifesaver. (Use Alipay to purchase your tickets.) You need to load the app on your phone away from China or with a VPN in order to get the English version.

Apple Translate

You can use the Google translate app (with a VPN), but I usually wind up using the built in Apple Translate app on my iPhone. (Alipay has a translation app, too). You’ll need at least one of them. Point your camera/phone at text on a sign or menu for translation, or use ‘conversation” mode to translate speech as text, or even a spoken conversation.

Trip

For booking flights and high speed trains in China, use Trip. They are reliable and pretty much cover all options. Trains go almost anywhere, but domestic flights are cheap and plentiful. You can book both with Trip. Also Trip is the easiest for booking hotels in China. Just out of habit I still use Booking.com, which also works for hotels (not trains or planes). Trip has the best selection in less touristy places, while Booking.com will work for most cities.

每周链接，2025年10月3日 || Weekly Links, 10/03/2025

2025-10-08T22:21:21+00:00

AI代理喜欢发帖，如果他们写博客会表现得更好：我们为代理构建了社交媒体，他们不会停止发帖

---------------

AI agents like to post and do better if they blog: We Built Social Media for Agents and They Won’t Stop Posting

支付给AI让我阅读我的书 || Paying AIs to Read My Books

2025-10-07T17:09:36+00:00

一些作者的观点颠倒了。他们认为AI公司应该为在他们的书籍上训练AI而支付报酬。但我的预测是，在不久的将来，作者们将支付给AI公司，以确保他们的书籍被纳入AI的教育和训练中。作者（以及他们的出版商）会支付费用，以便对AI提供的答案和服务产生影响。如果您的作品不被AI所知悉和欣赏，那么它实际上将无人知晓。

最近，AI公司Anthropic同意向书籍作者支付总计15亿美元的集体罚金，作为非法复制其书籍的惩罚。Anthropic曾因使用包含50万本书籍的“影子图书馆”而被一些作者起诉，该图书馆包含了这些作者书籍的数字版本，均由一些梦想让所有书籍对所有人开放的“叛逆图书管理员”收集。Anthropic曾下载了这个非法图书馆的副本，以备用于训练其大语言模型（LLMs），但根据法院文件，他们最终并未使用这些书籍来训练他们发布的人工智能模型。即使Anthropic没有使用这个特定的图书馆，他们也使用了类似的东西，而所有其他商业前沿的大语言模型也都是如此。

然而，法官处罚他们是因为未经授权复制了受版权保护的书籍，无论是否实际使用了这些书籍，所有被复制书籍的作者都获得了每本书3000美元的赔偿。

在本案中，法院管理员发布了受影响书籍的可搜索列表，专门网站上列出了这些书籍。任何人都可以搜索数据库，查看特定书籍或作者是否包含在该盗版图书馆中，当然也可以确认他们是否应获得赔偿。我过去参与过类似的集体诉讼案件，发现很少有实际的钱能真正到达普通民众手中。大部分费用都被各方律师消耗掉了。我注意到，在本案中，每本书的赔偿金额中只有50%会真正支付给作者，而另一半则归出版商所有。也许如此。如果是教科书，你可能连一点补偿都得不到。

我是一名作者，因此我查看了Anthropic的案件列表。我发现我出版的五本书中有四本被包含在该图书馆中。我感到荣幸能被纳入一组可以训练AI的书籍中，这些AI我现在每天都在使用。我也感到自豪，因为我的想法可能通过LLMs的思维链影响到数以百万计的人。我也可以想象一些作者会感到失望，因为他们的作品没有被包含在该图书馆中。

然而，Anthropic声称他们并未使用这个特定的图书馆来训练他们的AI。他们可能使用了其他图书馆，而这些图书馆可能或可能没有合法地获得授权。使用数字化书籍的合法性仍存在争议。例如，Google为了搜索目的数字化书籍，但只显示书籍的片段作为搜索结果。他们是否可以使用已经制作的数字副本用于训练AI？在Bartz诉Anthropic一案中，裁决认为，如果书籍副本是通过正当方式获得的，那么用于训练AI的书籍副本属于合理使用。Anthropic被处罚的原因不是因为用书籍训练AI，而是因为他们持有未经付费的书籍副本。

这仅仅是关于AI训练的法律测试案例之一，但显然未来会有更多类似的测试。因为很明显，版权法无法覆盖这种新的文本使用方式。保护文本的副本——这是版权条款所做的事情——与学习和训练无关。AI不需要保存副本，它们只需要读一遍。副本并不重要。我们可能需要其他类型的知识产权权利和许可，例如“引用权”或类似的概念。但权利问题只是次要的，主要事件是AI这一新受众的崛起。

我们将逐渐积累一些最佳实践，以确定如何训练和教育AI。用于教育AI代理人的材料的整理将成为决定我们是否使用和依赖它们的重要因素。会有少数客户希望AI的训练材料符合他们的政治倾向。虔诚的保守派可能希望AI接受保守教育，这样它们在回答有争议的问题时会按照他们的喜好来呈现。虔诚的自由派则希望AI接受自由派教育。大多数人不会在意，他们只是想要“最好的”答案或最可靠的服务。我们知道AI会反映它们所训练的内容，并且可以通过人类干预进行“微调”，以产生符合用户喜好的答案和服务。在强化AI行为和引导其思维方面，已有大量研究。

50万本书听起来像是很多书籍可供学习，但世界上已有数以百万计的书籍未被AI阅读，因为它们的版权状态不明确或不便，或者它们是用较少使用的语言书写的。AI的训练远未完成。塑造这一可能影响的语料库将本身成为一门科学和艺术。某一天，AI将真正阅读人类写下的所有内容。目前只有50万本书构成你的知识库，很快就会显得很陈旧，但这也表明，被包含在这一小部分书籍中将有多么大的影响力，这正是作者们现在希望他们的作品被AI训练的原因。

年轻人和最早采用AI的人群已经将AI设置为始终开启模式；越来越多的无形生活通过AI进行，且不再有其他途径。随着AI模型变得越来越可靠，年轻人开始接受AI的结论。我在自己的生活中也发现了类似的现象。我很久以前就不再质疑计算器，然后也不再质疑谷歌，现在我发现当前的AI提供的大多数答案都很可靠。AI正在成为真理的仲裁者。

AI代理不仅被用来提供答案，还被用来查找事物、理解事物和提出建议。如果AI不知道某件事情，那就相当于它不存在。因此，选择退出AI训练的作者将很难产生影响。目前，有相当多的作者和创作者根本没有数字存在；你无法在网上找到他们，他们的作品也没有被任何地方列出。他们非常稀少，属于少数群体。正如Tim O’Reilly所说，如今大多数创作者面临的挑战不是盗版（非法复制），而是被忽视。我补充说，未来创作者面临的挑战将不是模仿（AI复制），而是被忽视。

如果AI成为真理的仲裁者，并且它们所训练的内容至关重要，那么我希望我的想法和创作作品在它们所看到的内容中占据主导地位。我非常希望我的书籍成为AI的教科书。哪个作者不想呢？我也不例外。我希望我的影响力扩展到每天使用AI的数十亿人，我甚至可能愿意为此付费，或者至少做些事情来促进我的作品被AI吸收。

另一种思考方式是，在这个新兴的环境中，书籍的受众——尤其是非虚构类书籍——已经从人类转向了AI。如果你今天正在写一本书，你必须意识到你主要是为AI写作。它们将是阅读最多、最仔细的读者。它们会逐字逐句地阅读每一本书，包括脚注、尾注、参考文献和附录。它们还会阅读你所有的书籍并收听你的播客。你不太可能有任何人比AI更彻底地阅读你的作品。在吸收了这些内容后，AI会做一件神奇的事情：将你的文本与其他它们读过的文本结合，将它置于全球知识体系中，以一种人类读者无法做到的方式。

被AI吸收的材料的成功部分取决于这些材料如何被呈现给AI。如果一本书更容易被AI解析，它的影响力就会更大。因此，许多书籍将被撰写和格式化，以适应其主要受众。为AI写作将成为一种技能，就像其他任何技能一样，你也可以通过练习来提高。作者们可能会积极寻求优化他们的作品以适应AI的吸收，甚至可能与AI公司合作，确保他们的内容被正确理解并整合。这种“AI友好型”写作，具有清晰的结构、明确的论点和定义良好的概念，将变得越来越重要，当然，AI本身也会帮助实现这一点。

我们创作的每一本书、歌曲、戏剧、电影都会被加入我们的文化。图书馆是人类发明中特别的一种。它们往往越老越有价值。它们积累智慧和知识。互联网在这方面也类似，它持续积累材料，自诞生以来从未崩溃或需要重启。AI很可能也类似这些外向系统，不断积累知识而不会中断。我们尚不确定，但它们很可能在数十年甚至更长时间内持续增长。目前它们的增长似乎是无止境的。今天学到的知识，它们明天仍然会记得，而它们今天的影响将在未来的几十年中产生复利效应。影响AI将成为当今人类可进行的最高杠杆效应活动，而且你开始得越早，效果就越显著。

作者作品的价值将不仅体现在它在人类中的销售情况，还体现在它被包含在这些基于记忆的智能系统的基础知识中的深度。这种影响力将成为宣传的重点，也将成为作者的遗产。

---------------

Some authors have it backwards. They believe that AI companies should pay them for training AIs on their books. But I predict in a very short while, authors will be paying AI companies to ensure that their books are included in the education and training of AIs. The authors (and their publishers) will pay in order to have influence on the answers and services the AIs provide. If your work is not known and appreciated by the AIs, it will be essentially unknown.

Recently, the AI firm Anthropic agreed to pay book authors a collective $1.5 billion as a penalty for making an illegal copy of their books. Anthropic had been sued by some authors for using a shadow library of 500,000 books that contained digital versions of their books, all collected by renegade librarians with the dream of making all books available to all people. Anthropic had downloaded a copy of this outlaw library in anticipation of using it to train their LLMs, but according to court documents, they did not end up using those books for training the AI models they released. Even if Anthropic did not use this particular library, they used something similar, and so have all the other commercial frontier LLMs.

However the judge penalized them for making an unauthorized copy of the copyrighted books, whether or not they used them, and the authors of all the copied books were awarded $3,000 per book in the library.

The court administrators in this case, called Bartz et al v. Anthropic, have released a searchable list of the affected books on a dedicated website. Anyone can search the database to see if a particular book or author is included in this pirate library, and of course, whether they are due compensation. My experience with class action suites like this is that very rarely does award money ever reach people on the street. Most of the fees are consumed by the lawyers of all sides. I notice that in this case, only half of the amount paid per book is destined to actually go to the author. The other 50% goes to the publishers. Maybe. And if it is a text book, good luck with getting anything.

I am an author so I checked the Anthropic case list. I found four out of my five books published in New York included in this library. I feel honored to be included in a group of books that can train AIs that I now use everyday. I feel flattered that my ideas might be able to reach millions of people through the chain of thought of LLMs. I can imagine some authors feeling disappointed that their work was not included in this library.

However, Anthropic claims it did not use this particular library for training their AIs. They may have used other libraries and those libraries may or may not have been “legal” in the sense of having been paid for. The legality of using digitized books for anything is still in dispute. For example, Google digitizes books for search purposes, but only shows small snippets of the book as the result. Can they use the same digital copy they have already made for training AI purposes? The verdict in the Bartz v. Anthropic case was that, yes, using a copy of a book for training AI is fair use, if it was obtained in a fair way. Anthropic was penalized not for training AI on books, but for having in its possession a copy of the books it had not paid for.

This is just the first test case of what promises to be many more tests in the future as it is clear that copyright law is not adequate to cover this new use of text. Protecting copies of text – which is what copyright provisions do – is not really pertinent to learning and training. AIs don’t need to keep a copy; they just have to read it once. Copies are immaterial. We probably need other types of rights and licenses for intellectual property, such as a Right of Reference, or something like that. But the rights issue is only a distraction from the main event, which is the rise of a new audience: the AIs.

Slowly, we’ll accumulate some best practices in regards to what is used to train and school AIs. The curation of the material used to educate the AI agents giving us answers will become a major factor in deciding whether we use and rely on them. There will be a minority of customers who want the AIs to be trained with material that aligns with their political bent. Devout conservatives might want a conservatively trained AI; it will give answers to controversial questions in the manner they like. Devout liberals will want one trained with a liberal education. The majority of people won’t care; they just want the “best” answer or the most reliable service. We do know that AIs reflect what they were trained on, and that they can be “fine tuned” with human intervention to produce answers and services that please their users. There is a lot of research in reinforcing their behavior and steering their thinking.

Half a million books sounds like a lot of books to learn from, but there are millions and millions of books in the world already that the AIs have not read because their copyright status is unclear or inconvenient, or they are written in lesser-used languages. AI training is nowhere near done. Shaping this corpus of possible influences will become a science and art in itself. Someday AIs will have really read all that humans have written. Having only 500,000 books forming your knowledge base will soon be seen as quaint, but it also suggests how impactful it can be to be included in that small selection, and that makes inclusion a prime reason why authors will want their works to be trained on AIs now.

The young and the earliest adopters of AI have it set to always-on mode; more and more of their intangible life goes through the AI, and no further. As the AI models become more and more reliable, the young are accepting the conclusions of the AI. I find something similar in my own life. I long ago stopped questioning a calculator, then stopped questioning Google, and now find that most answers from current AIs are pretty reliable. The AIs are becoming the arbiters of truth.

AI agents are used not just to give answers but to find things, to understand things, to suggest things. If the AIs do not know about it, it is equivalent to it not existing. It will become very hard for authors who opt out of AI training to make a dent. There are authors and creators today who do not have any digital presence at all; you cannot find them online; their work is not listed anywhere. They are rare and a minority. As Tim O’Reilly likes to say, the challenge today for most creators is not piracy (illegal copies) but obscurity. I will add, the challenge for creators in the future will not be imitation (AI copy) but obscurity.

If AIs become the arbiters of truth, and if what they trained on matters, then I want my ideas and creative work to be paramount in what they see. I would very much like my books to be the textbooks for AI. What author would not? I would. I want my influence to extend to the billions of people coming to the AIs everyday, and I might even be willing to pay for that, or to at least do what I can to facilitate the ingestion of my work into the AI minds.

Another way to think of this is that in this emerging landscape, the audience for books – especially non-fiction books – has shifted away from people towards AI. If you are writing a book today, you want to keep in mind that you are primarily writing it for AIs. They are the ones who are going to read it the most carefully. They are going to read every page word by word, and all the footnotes, and all the endnotes, and the bibliography, and the afterward. They will also read all your books and listen to all your podcasts. You are unlikely to have any human reader read it as thoroughly as the AIs will. After absorbing it, the AIs will do that magical thing of incorporating your text into all the other text they have read, of situating it, of placing it among all the other knowledge of the world – in a way no human reader can do.

Part of the success of being incorporated by AIs is how well the material is presented for them. If a book can be more easily parsed by an AI, its influence will be greater. Therefore many books will be written and formatted with an eye on their main audience. Writing for AIs will become a skill like any other, and something you can get better at. Authors could actively seek to optimize their work for AI ingestion, perhaps even collaborating with AI companies to ensure their content is properly understood, and integrated. The concept of “AI-friendly” writing, with clear structures, explicit arguments, and well-defined concepts, will gain prominence, and of course will be assisted by AI.

Every book, song, play, movie we create is added to our culture. Libraries are special among human inventions. They tend to get better the older they get. They accumulate wisdom and knowledge. The internet is similar in this way, in that it keeps accumulating material and has never crashed, or had to restart, since it began. AIs are very likely similar to these exotropic systems, accumulating endlessly without interruption. We don’t know for sure, but they are liable to keep growing for decades if not longer. At the moment their growth seems open ended. What they learn today, they will probably continue to know, and their impact today will have compounding influence in the decades to come. Influencing AIs is among the highest leverage activities available to any human being today, and the earlier you start, the more potent.

The value of an author’s work will not just be in how well it sells among humans, but how deep it has been included within the foundational knowledge of these intelligent memory-based systems. That potency will be what is boasted about. That will be an author’s legacy.

认知的周期表 || The Periodic Table of Cognition

2025-09-23T22:50:22+00:00

我一直在研究电的发现早期历史，作为我们当前发现人造智能的指南。当时最聪明的人，包括艾萨克·牛顿，他可能是有史以来最聪明的人，都对电的本质有自信的理论，但这些理论却严重错误。事实上，尽管电荷在宇宙中起着基本作用，但很长一段时间内，所有研究这一基本力量的人都严重错误。所有电的先驱者——如富兰克林、惠斯通、法拉第和麦克斯韦——都有他们自己的正确想法（并非所有人都有），但这些想法中混杂着许多后来被证明是完全误导的观念。关于电能做什么的大多数发现，都是在不了解其工作原理的情况下进行的。这种知识的缺乏，当然，极大地减缓了电气发明的进步。同样地，今天最聪明的人，尤其是所有创造人工智能的天才，都有关于智能本质的理论，我认为所有这些理论（包括我自己）都会严重错误。我们之所以不知道人工智能是什么，很大程度上是因为我们不知道自己的智能是什么。这种无知将来会被视为阻碍人工智能进步速度的因素。我们对智能的无知很大程度上源于我们对电或智能这一总体类别的混淆。我们倾向于将电和智能视为单一维度上的连贯基本力量：你拥有更多或更少。但事实上，电被证明是如此复杂，如此充满反直觉效应，以至于即使今天，我们仍然难以理解它的工作原理。电有粒子和波，有场和流，由并不存在的事物组成。我们对电的使用超过了我们对其的理解。理解电对于理解物质至关重要。直到我们学会了控制电，才得以将水——曾被视为元素——分解为它的实际元素；这使我们认识到水并不是基础元素，而是一种由子元素组成的衍生化合物。我认为我们很可能也会发现，智能同样不是一种基础的单一元素，而是一种由多种认知元素组成的衍生化合物，每种智能物种都有其独特的复杂系统。我们称之为智能的结果，来源于许多不同的认知原始元素，如长期记忆、空间意识、逻辑推理、预先规划、模式感知等。可能有几十种，甚至上百种。我们目前对这些元素一无所知。我们缺乏认知的元素周期表。这些认知元素更像重元素一样不稳定且动态。或者更好的类比是生物细胞中的元素。认知的原始元素是思维循环中的流动状态。它们就像细胞中的分子，处于持续的流动中，不断改变形状。它们的分子身份与其与其他分子的作用和互动有关。思维是一种集体行为，发生在时间中（就像物质中的温度），每种模式只能在与其他模式之前和之后的关系中看到。这是一种网络现象，使得确定其边界变得困难。因此，每个智能元素都嵌入在思维循环中，并需要其他元素作为其身份的一部分。因此，每个认知元素都是在与其相邻的认知模式的背景下描述的。我让ChatGPT5Pro帮助我生成一个认知元素周期表，基于我们目前所知的信息。它建议有49个元素，排列成一个表格，使得相关概念相邻。列是认知的家族或一般类别，如“感知”、“推理”、“学习”，因此所有类型的感知或推理都堆叠在某一列中。行按思维循环的阶段排序，早期阶段（如“感知”）位于顶部，而循环后期阶段（如“反思与对齐”）位于底部。例如，在“安全”这一家族或类别中，人工智能会先进行不确定性估计，之后进行验证，最后才达到心智理论。该图表根据我们在每个元素上取得的进展进行着色。红色表示我们能够以稳健的方式合成该元素。橙色表示我们可以在正确框架下使其大致运行。黄色则反映有前景的研究，但尚未具备操作上的普遍性。我怀疑这些元素之间的区分可能不像这里所展示的那样明显（在分类学上，我更倾向于合并而非细分），并且我认为这个集合可能遗漏了许多我们即将发现的类型。但作为起点，这个原型图表达到了它的目的：它揭示了智能的复杂性。很明显，智能是在多个维度上叠加的。我们将设计不同的AI，使其具有不同元素的不同组合和不同强度，这将产生数千种可能的智能类型。我们可以看到，即使今天，不同动物也有自己的认知原始元素组合，这些组合以符合其物种需求的独特模式排列。在某些动物中，某些元素——比如长期记忆——可能比我们自己的更强；当然，它们缺乏我们拥有的某些元素。
---------------

I’ve been studying the early history of electricity’s discovery as a map for our current discovery of artificial intelligence. The smartest people alive back then, including Isaac Newton, who may have been the smartest person who ever lived, had confident theories about electricity’s nature that were profoundly wrong. In fact, despite the essential role of electrical charges in the universe, everyone who worked on this fundamental force was profoundly wrong for a long time. All the pioneers of electricity — such as Franklin, Wheatstone, Faraday, and Maxwell — had a few correct ideas of their own (not shared by all) mixed in with notions that mostly turned out to be flat out misguided. Most of the discoveries about what electricity could do happened without the knowledge of how they worked. That ignorance, of course, drastically slowed down the advances in electrical inventions.

In a similar way, the smartest people today, especially all the geniuses creating artificial intelligence, have theories about what intelligence is, and I believe all of them (me too) will be profoundly wrong. We don’t know what artificial intelligence is in large part because we don’t know what our own intelligence is. And this ignorance will later be seen as an impediment to the rate of progress in AI.

A major part of our ignorance stems from our confusion about the general category of either electricity or intelligence. We tend to view both electricity and intelligence as coherent elemental forces along a single dimension: you either have more of it or less. But in fact, electricity turned out to be so complicated, so complex, so full of counterintuitive effects that even today it is still hard to grasp how it works. It has particles and waves, and fields and flows, composed of things that are not really there. Our employment of electricity exceeds our understanding of it. Understanding electricity was essential to understanding matter. It wasn’t until we learned to control electricity that we were able to split water — which had been considered an element — into its actual elements; that enlightened us that water was not a foundational element, but a derivative compound made up of sub elements.

It is very probable we will discover that intelligence is likewise not a foundational singular element, but a derivative compound composed of multiple cognitive elements, combined in a complex system unique to each species of mind. The result that we call intelligence emerges from many different cognitive primitives such as long-term memory, spatial awareness, logical deduction, advance planning, pattern perception, and so on. There may be dozens of them, or hundreds. We currently don’t have any idea of what these elements are. We lack a periodic table of cognition.

The cognitive elements will more resemble the heavier elements in being unstable and dynamic. Or a better analogy would be to the elements in a biological cell. The primitives of cognition are flow states that appear in a thought cycle. They are like molecules in a cell which are in constant flux, shifting from one shape to another. Their molecular identity is related to their actions and interactions with other molecules. Thinking is a collective action that happens in time (like temperature in matter) and every mode can only be seen in relation to the other modes before and after it. It is a network phenomenon that makes it difficult to identify its borders. So each element of intelligence is embedded in a thought cycle, and requires the other elements as part of its identity. So each cognitive element is described in context of the other cognitive modes adjacent to it.

I asked ChatGPT5Pro to help me generate a periodic table of cognition given what we collectively know so far. It suggests 49 elements, arranged in a table so that related concepts are adjacent. The columns are families, or general categories of cognition such as “Perception”, “Reasoning”, “Learning”, so all the types of perception or reasoning are stacked in one column. The rows are sorted by stages in a cycle of thought. The earlier stages (such as “sensing”) are at the top, while later stages in the cycle (such as “reflect & align”) are at the bottom. So for example, in the family or category of “Safety” the AIs will tend to do the estimation of uncertainty first, later do verification, and only get to a theory of mind at the end.

The chart is colored according to how much progress we’ve made on each element. Red indicates we can synthesize that element in a robust way. Orange means we can kind of make it work with the right scaffolding. Yellow reflects promising research without operational generality yet.

I suspect many of these elements are not as distinct as shown here (taxonomically I am more of a lumper than a splitter), and I would expect this collection omits many types we are soon to discover, but as a start, this prototype chart serves its purpose: it reveals the complexity of intelligence. It is clear intelligence is compounded along multiple dimensions. We will engineer different AIs to have different combinations of different elements in different strengths. This will produce thousands of types of possible minds. We can see that even today different animals have their own combination of cognitive primitives, arranged in a pattern unique to their species’ needs. In some animals some of the elements — say long-term memory — may exceed our own in strength; of course they lack some elements we have.

With the help of AI, we are discovering what these elements of cognition are. Each advance illuminates a bit of how minds work and what is needed to achieve results. If the discovery of electricity and atoms has anything to teach us now, it is that we are probably very far from having discovered the complete set of cognitive elements. Instead we are at the stage of believing in ethers, instantaneous action, and phlogiston – a few of the incorrect theories of electricity the brightest scientists believed.

Almost no thinker, researcher, experimenter, or scientist at that time could see the true nature of electricity, electromagnetism, radiation and subatomic particles, because the whole picture was hugely unintuitive. Waves, force fields, particles of atoms did not make sense (and still does not make common sense). It required sophisticated mathematics to truly comprehend it, and even after Maxwell described it mathematically, he found it hard to visualize.

I expect the same from intelligence. Even after we identify its ingredients, the emergent properties they generate are likely to be obscure and hard to believe, hard to visualize. Intelligence is unlikely to make common sense.

A century ago, our use of electricity ran ahead of our understanding of it. We made motors from magnets and coiled wire without understanding why they worked. Theory lagged behind practice. As with electricity, our employment of intelligence exceeds our understanding of it. We are using LLMs to answer questions or to code software without having a theory of intelligence. A real theory of intelligence is so lacking that we don’t know how our own minds work, let alone the synthetic ones we can now create.

The theory of the atomic world needed the knowledge of the periodic table of elements. You had to know all (or at least most) of the parts to make falsifiable predictions of what would happen. The theory of intelligence requires knowledge of all the elemental parts, which we have only slowly begun to identify, before we can predict what might happen next.

信任指数 (TQ) || The Trust Quotient (TQ)

2025-09-02T20:15:23+00:00

无论何处存在自主性，信任都必须随之而来。如果我们培养孩子独立生活，他们需要具备自主性，而我们也需要信任他们。（育儿是学习信任的学校。）如果我们创建一个由自主代理组成的系统，就需要在代理之间建立大量的信任。如果我将决策委托给AI，那么我必须信任它；如果该AI依赖其他AI，它也必须信任它们。因此，我们需要开发一个非常稳健的信任系统，能够检测、验证和在人类与机器之间，更重要的是在机器与机器之间建立信任。关于信任的研究主要沿着两个方向展开：更好地理解人类之间如何建立信任，以及将这些原则以抽象的方式应用于机械系统中。技术专家已经创建了初步的信任系统来管理数据云和通信的安全。例如，这个设备是否应该被允许连接？它是否值得信赖以完成它声称能做的事情？我们如何验证其身份和行为？等等。目前这些系统并未处理适应性代理，它们的行为、身份和能力远比现有系统更加多变、模糊、不稳定，而且影响更为深远。这使得信任它们变得更加困难和重要。如今，当我寻找一个AI时，准确度是我最看重的品质。它是否能给出正确的答案？它会有多大的幻觉？这些品质是信任的代理。我可以信任这个AI给出可靠的答案吗？随着AI开始承担更多任务，进入现实世界进行行动和决策，它们的可信度变得至关重要。信任是一个广泛的概念，随着它渗入AI生态系统，它将被拆解为更具体的组成部分。作为安全、可靠性、责任和问责的一部分，这些要素将随着我们对其的综合和测量而变得更加精确。在未来十年中，信任将成为我们讨论得更多的话题。随着AI的能力和技能开始分化——有些更适合特定任务——对它们的评价将开始包括其可信度。就像其他制造产品有广告的规格——如燃油效率、存储容量、像素数量、可用时间或治愈率——AI的供应商也将开始宣传其代理的“信任指数”。它们有多可靠？即使这种品质未被宣传，也需要在内部进行测量，以便公司能够不断改进它。当我们依赖AI代理预订旅行票、续订药品处方或修理汽车时，我们将对其寄予厚望。很难想象AI代理会参与生死攸关的决定。我们甚至可能面临法律后果，因为对AI代理的信任程度。如果代理出了差错，谁来负责？目前，AI并不承担任何责任。如果它们犯了错误，它们不保证修复。它们不对因错误可能引发的问题负责。事实上，这种差异目前是人类员工与AI员工之间的关键区别。责任最终由人类承担，他们负责自己的工作；你雇佣人类是因为你信任他们能正确完成工作。如果未能做到，他们会重新做，并学习如何不再犯同样的错误。而当前的AI则并非如此，这使得它们难以信任。 AI代理将形成一个网络，一个相互作用的AI系统，该系统可以为每个任务分配风险因素。一些任务，如购买机票或开具处方药，会有风险评分反映潜在的负面结果与正面便利。每个AI代理本身也会根据其权限拥有动态的风险评分。代理们还会根据过去的绩效积累信任评分。信任是非常不对称的；它可能需要许多互动和长时间才能积累价值，但一旦犯错，信任可能瞬间丧失。信任评分将不断变化，并由系统追踪。大多数AI工作将通过代理之间的交流完成，而不被人类察觉。大多数由普通AI代理生成的输出将仅被另一个AI代理所消费，数量庞大。整个AI工作的极少数部分将被人类看到或注意到。人类与之互动的AI代理数量将非常少，尽管它们对我们来说至关重要。虽然我们接触的AI代理在统计上是罕见的，但它们对我们意义重大，其信任度将成为关键。为了赢得我们的信任，一个面向外部的AI代理需要连接到它也能信任的AI代理，因此其能力的很大一部分将是选择和利用最值得信赖的AI代理的技能。我们可以预见全新的骗局，包括欺骗AI代理信任空壳代理，伪造信任证书，假冒身份，伪装任务。就像在互联网安全领域，一个AI代理的可信度取决于其最薄弱的子代理。由于子任务可以分配到多个层级，管理质量将成为AI的重要任务。对错误的正确归因和纠正错误也将成为AI的巨大可销售技能。所有系统——包括最优秀的人类——都会犯错。没有完全无误的系统。因此，高信任度的很大一部分在于对错误负责并加以纠正。最受信任的代理将是那些能够（并且被信任！）修复自己所犯的错误，拥有足够的智能能力进行弥补并确保正确性的代理。最终，我们给予主要AI代理的信任程度——也就是我们每天与之互动的代理——将是一个被夸耀、争论、共享和广泛宣传的评分。在其他领域，如汽车或手机，我们默认其可靠性。 AI比其他产品和服务更加复杂和个性化，不同于我们今天生活中的其他产品和服务， AI代理的可信度将成为关键，并且是一个持续的关注点。其信任指数（TQ）可能比其智力指数（IQ）更重要。选择和保留高TQ的代理将非常类似于招聘和留住关键的人类员工。然而，我们倾向于避免给人类分配数值评分。另一方面，AI代理系统将拥有各种我们用来决定是否希望它们帮助管理我们生活的指标。评分最高的AI可能也是最昂贵的。会有传言称那些几乎完美评分的AI你负担不起。然而，AI是一个随着使用而不断改进的系统，这意味着使用越多，它变得越好，因此最好的AI将是最受欢迎的。亿万富翁使用与我们相同的谷歌，很可能也会使用与我们相同的AI，尽管他们可能拥有高度个性化的界面。这些同样需要拥有最高的信任指数。
---------------

Wherever there is autonomy, trust must follow. If we raise children to go off on their own, they need to be autonomous and we need to trust them. (Parenting is a school for learning how to trust.) If we make a system of autonomous agents, we need lots of trust between agents. If I delegate decisions to an AI, I then have to trust it, and if that AI relies on other AIs, it must trust them. Therefore we will need to develop a very robust trust system that can detect, verify, and generate trust between humans and machines, and more importantly between machines and machines.

Applicable research in trust follows two directions: understanding better how humans trust each other, and applying some of those principles in an abstract way into mechanical systems. Technologists have already created primitive trust systems to manage the security of data clouds and communications. For instance, should this device be allowed to connect? Can it be trusted to do what it claims it can do? How do we verify its identity, and its behavior? And so on.

So far these systems are not dealing with adaptive agents, whose behaviors and IDs and abilities are far more fluid, opaque, shifting, and also more consequential. That makes trusting them more difficult and more important.

Today when I am shopping for an AI, accuracy is the primary quality I am looking for. Will it give me correct answers? How much does it hallucinate? These qualities are proxies for trust. Can I trust the AI to give me an answer that is reliable? As AIs start to do more, to go out into the world to act, to make decisions for us, their trustworthiness becomes crucial.

Trust is a broad word that will be unbundled as it seeps into the AI ecosystem. Part security, part reliability, part responsibility, and part accountability, these strands will become more precise as we synthesize it and measure it. Trust will be something we’ll be talking a lot more about in the coming decade.

As the abilities and skills of AI begin to differentiate – some are better for certain tasks than others – reviews of them will begin to include their trustworthiness. Just as other manufactured products have specs that are advertised – such as fuel efficiency, or gigabytes of storage, pixel counts, or uptime, or cure rates – so the vendors of AIs will come to advertise the trust quotient of their agents. How reliably reliable are they? Even if this quality is not advertised it needs to be measured internally, so that the company can keep improving it.

When we depend on our AI agent to book vacation tickets, or renew our drug prescriptions, or to get our car repaired, we will be placing a lot of trust in them. It is not hard to imagine occasions where an AI agent can be involved in a life or death decision. There may even be legal liability consequences for how much we can expect to trust AI agents. Who is responsible if the agent screws up?

Right now, AIs own no responsibilities. If they get things wrong, they don’t guarantee to fix it. They take no responsibility for the trouble they may cause with their errors. In fact, this difference is currently the key difference between human employees and AI workers. The buck stops with the humans. They take responsibility for their work; you hire humans because you trust them to get the job done right. If it isn’t, they redo it, and they learn how to not make that mistake again. Not so with current AIs. This makes them hard to trust.

AI agents will form a network, a system of interacting AIs, and that system can assign a risk factor for each task. Some tasks, like purchasing airline tickets, or assigning prescription drugs, would have risk scores reflecting potential negative outcomes vs positive convenience. Each AI agent itself would have a dynamic risk score depending on what its permissions were. Agents would also accumulate trust scores based on their past performances. Trust is very asymmetrical; It can take many interactions over a long time to gain in value, but it can lose trust instantly, with a single mistake. The trust scores would be constantly changing, and tracked by the system.

Most AI work will be done invisibly, as agent to agent exchanges. Most of the output generated by an average AI agent will only be seen and consumed by another AI agent, one of trillions. Very little of the total AI work will ever be seen or noticed by humans. The number of AI agents that humans interact with will be very few, although they will loom in importance to us. While the AIs we engage with will be rare statistically, they will matter to us greatly, and their trust will be paramount.

In order to win that trust from us, an outward facing AI agent needs to connect with AI agents it can also trust, so a large part of its capabilities will be the skill of selecting and exploiting the most trustworthy AIs it can find. We can expect whole new scams, including fooling AI agents into trusting hollow agents, faking certificates of trust, counterfeiting IDs, spoofing tasks. Just as in the internet security world, an AI agent is only as trustworthy as its weakest sub-agent. And since sub-tasks can be assigned for many levels down, managing quality will be a prime effort for AIs.

Assigning correct blame for errors and rectifying mistakes also becomes a huge marketable skill for AIs. All systems – including the best humans – make mistakes. There can be no system mistake proof. So a large part of high trust is the accountability in mending one’s errors. The highest trusted agents will be those capable (and trusted!) to fix the mistakes they make, to have sufficient smart power to make amends, and get it right.

Ultimately the degree of trust we give to our prime AI agent — the one we interact with all day every day — will be a score that is boasted about, contested, shared, and advertised widely. In other domains, like a car or a phone, we take reliability for granted.

AI is so much more complex and personal, unlike other products and services in our lives today,

the trustworthiness of AI agents will be crucial and an ongoing concern. Its trust quotient (TQ) may be more important than its intelligence quotient (IQ). Picking and retaining agents with high TQ will be very much like hiring and keeping key human employees.

However, we tend to avoid assigning numerical scores to humans. The AI agent system, on the other hand will have all kinds of metrics we will use to decide which ones we want to help run our lives. The highest scoring AIs will likely be the most expensive ones as well. There will be whispers of ones with nearly perfect scores that you can’t afford. However, AI is a system that improves with increasing returns, which means the more it is used, the better it gets, so the best AIs will be among the most popular AIs. Billionaires use the same Google we use, and are likely to use the same AIs as us, though they might have intensely personalized interfaces for them. These too, will need to have the highest trust quotients.

Every company, and probably every person, will have an AI agent that represents them inside the AI system to other AI agents. Making sure your personal rep agent has a high trust score will be part of your responsibility. It is a little bit like a credit score for AI agents. You will want a high TQ for yours. Because some AI agents won’t engage with other agents having low TQs. This is not the same thing as having a personal social score (like the Chinese are reputed to have). This is not your score, but the TQ score of your agent, which represents you to other agents. You could have a robust social score reputation, but your agent could be lousy. And vice versa.

In the coming decades of the AI era, TQ will be seen as more important than IQ.

情感代理 || Emotional Agents

2025-08-23T21:57:26+00:00

许多人发现人工智能的智能令人震惊。然而，与即将到来的更大冲击相比，这似乎微不足道：高度情感化的人工智能。合成情感的出现将在人类社会引发颠覆、愤怒、困扰、混乱和文化冲击，其影响将远超对合成智能的讨论。在接下来的几年里，新闻标题将从“所有人都会失业”（其实不会）转变为“人工智能伴侣是人类文明的终结”。我们能够理性地接受计算机可能具备理性。尽管我们可能不喜欢，但我们可以接受计算机可能变得聪明，部分原因是我们已经将大脑视为一种计算机。很难相信它们能像我们一样聪明，但一旦它们做到了，这似乎也合乎逻辑。接受机器制造的创造力则更为困难。创造力似乎非常人类化，某种程度上被视为理性的对立面，因此看起来并不属于机器。情感是有趣的，因为情感不仅存在于人类，还存在于许多动物身上。任何宠物主人都能列举出宠物如何感知和表达情感的方式。人们喜爱动物的部分原因在于能够与它们在情感上产生共鸣。它们对我们的感情作出回应，就像我们对它们的感情作出回应一样。人类与动物之间存在真实而深刻的情感纽带。同样的情感纽带也将出现在机器上。我们已经看到一些迹象。几乎每周都有陌生人向我发送他们与AI聊天的日志，展示这些互动多么深刻和直觉，它们如何理解彼此，以及在精神上如何连接。我们还收到关于青少年与AI“朋友”建立深厚关系的报告。这一切都发生在尚未有系统性地将情感嵌入AI之前。我们为何要将情感编程进AI？有几个原因：首先，情感是与机器交互的绝佳界面。它使互动更加自然和舒适。情感对人类来说很容易。我们无需学习如何行动，我们都能直观地理解诸如表扬、热情、怀疑、说服、惊喜、困惑等结果，这些是机器可能希望使用的。人类通过微妙的情感色彩传达非语言信息、重要性以及指令，而AI将在其指令和交流中使用类似的情感线索。其次，市场将更青睐情感代理，因为人类本身也是如此。随着AI和机器人能力的趋同，它们的多样性将继续发展，因此它们的性格和情感特征在选择使用时将变得更为重要。如果它们都同样聪明，那么更友善、更亲切或更好的陪伴者将赢得用户的青睐。第三，我们希望人工智能代理完成的许多任务，无论是软件AI还是实体机器人，都需要的不仅仅是理性计算。仅仅让AI整夜编程是不够的。我们目前对智能的评价过高。要真正具备创造力和创新能力，要足够明智以提供良好建议，所需要的不仅仅是智商。这些机器人需要深入其软件中的复杂情感动态。这是否可能？是的。已有几十年的研究项目（如麻省理工学院的研究）致力于将情感提炼为可移植到机器中的属性。其中一些知识涉及如何在硬件上视觉化情感，就像我们通过自己的面部表情来表达一样。其他研究者则提取了我们通过声音传达情感的方式，甚至通过文本中的词语。最近，我们见证了AI开发者调整其代理的友好程度和“亲切感”，因为一些用户不喜欢它们的新性格，而另一些人则单纯不喜欢性格的变化。虽然我们可以编程性格和情感，但我们尚未明确知道哪些情感最适合特定任务。机器展示情感只是工作的一半。另一半是机器对人类情感的检测和理解。关系是双向的，为了真正成为情感代理，机器必须擅长捕捉你的情感。该领域已有大量研究，主要集中在面部识别上，不仅识别身份，还识别情绪状态。已有商业应用可以观察用户在键盘上的输入，检测他们是否抑郁或处于情绪压力下。未来，智能眼镜将不仅观察外界，还会回过头来观察你的面部表情以解析你的情绪。你是否困惑或欣喜？是否惊讶或感激？是否坚定或放松？目前，苹果的Vision Pro已经配备了朝后摄像头，用于追踪眼睛和微表情，如眨眼和眉毛上扬。当前的文本大型语言模型（LLM）在检测用户情绪状态上几乎没有尝试，除了从提示中的字母中推断出的信息。但技术上来说，这并不是巨大的飞跃。在接下来的几年里，将会有大量关于情感的实验。一些AI会简洁而逻辑性强，一些则会健谈且外向。一些AI会低声细语，只在你准备好倾听时才说话。一些人会更喜欢那些能让他们发笑的幽默、机智的AI。许多商业AI将被设计成你的最佳朋友。我们可能会认为这是成年人的可取之处，但对儿童来说却可能令人不安。事实上，关于AI与儿童的关系，有许多问题需要警惕，不仅仅是情感方面。但情感纽带将是儿童AI设计中的关键考虑因素。年幼的人类儿童已经能够与惰性娃娃和泰迪熊建立情感联系，并变得非常亲密。想象一下，如果一个泰迪熊能回应你的对话、以无限耐心与你玩耍，并且能映射你的情感。随着孩子的成长，他们可能永远不会放弃这个泰迪熊。因此，机器中情感的质量很可能会成为我们制定不同制度的领域，一种针对成年人，一种针对儿童。不同的规则、不同的期望、不同的法律、不同的商业模式等。但即使是成年人，也会对情感代理产生强烈依恋，就像电影《她》中那样。起初，社会会将那些沉迷于AI爱情的人视为有幻觉或精神不稳定。但正如大多数深爱狗或猫的人并非精神有问题，而是正常且富有同理心的人一样，大多数与AI和机器人建立亲密关系的人也会将这些纽带视为健康且能拓展自我的。关于与机器建立亲密关系的常见恐惧是，它们可能太友善、太聪明、太有耐心、太可用，比周围的人类更有效，因此人们会完全脱离人类关系。这种情况可能发生。我们很容易想象，有善意的人只会消费AI提供的“甜美易得的友谊”，就像他们被诱惑去只消费加工食品中的“甜美易得的卡路里”一样。对抗这种诱惑的最佳方法与快餐类似：教育和更好的选择。在这个新世界中成长的一部分，就是学会区分那些看似完美的关系与人类关系中混乱、困难和不完美的部分，并认识到后者的价值。要成为你自己最好的版本——无论你如何定义——都需要与人类共处！与其禁止AI关系（或快餐），不如对其进行适度管理，并保持理性看待。事实上，AI朋友、导师、教练或伴侣的“完美”行为可以成为绝佳的榜样。如果你身边有经过训练和调整的AI，它们是人类所能创造的最好版本，这将是一种极好的自我提升方式。普通人类往往拥有非常浅薄的道德观，矛盾的准则，并且容易被自己的基本欲望和处境所左右。从理论上讲，我们应该能够编程AI，使其具备比普通人类更好的道德和准则。同样，我们也可以设计AI成为比普通人类更好的朋友。拥有这些受过教育的AI可以帮助我们提升自己，成为更好的人类。与它们建立深厚关系的人，可能成为最适应社会和富有同理心的人。关于AI的情感并非真实，因为“机器人无法感受任何东西”的论点将被忽视。就像批评人工智能是人工的，因此不真实，因为它们不理解一样。这并不重要。我们并不真正理解“感受”意味着什么，甚至也不清楚“理解”意味着什么。这些术语和概念已经成为习惯，但已不再有用。AI能够做我们过去称为智能的事情，它们也将开始做我们过去称为情感的事情。最重要的是，人类与AI、机器人之间的关系将与任何其他人类关系一样真实和有意义。它们将是一种真实的关系。但AI或机器人所拥有的情感，尽管真实，却可能有所不同。真实但有所偏差。AI可以幽默，但它们的幽默感略有偏差，略有不同。它们会笑我们不会笑的事情。它们的幽默方式会逐渐改变我们自己的幽默观，就像它们下棋和打麻将的方式已经改变了我们下棋和打麻将的方式一样。AI是聪明的，但以非人类的方式。它们的情感性也将同样陌生，因为AI本质上是人工的外星生命。事实上，我们将通过观察它们的情感来更深入地理解情感的本质，而不是通过研究自己。机器中的情感不会一夜之间出现。情感将逐渐积累，这样我们就有时间引导它们。它们最初表现为礼貌、文明和友善。它们会轻易地赞美和奉承我们，也许过于轻易。核心关注点并非是我们的机器伴侣关系是否会亲密（它们会），也不是这些关系是否真实（它们是），也不是它们是否会取代人类关系（它们不会），而是你的情感代理为你服务的是谁？谁拥有它？它被优化用于什么？你能信任它不会操纵你吗？这些问题将在未来十年主导我们的讨论。显然，关于我们的最敏感数据，莫过于来自我们情感的信息。我们害怕什么？什么让我们真正快乐？什么让我们感到厌恶？什么会激发我们？在与我们的全天候代理互动多年后，该代理将拥有我们的完整档案。即使我们从未明确透露过最深层的恐惧、最珍视的愿望和最脆弱的时刻，它也能从我们的交流、问题和反应的情感倾向中了解一切。它将比我们自己更了解我们。这将在未来几十年成为常见的说法，既令人兴奋又令人恐惧：“我的AI代理比我更了解我自己。” 在许多情况下，这将是真实的。在最好的情况下，我们可以利用这个工具更好地认识自己。在最坏的情况下，这种知识的不对称性将被用来操纵我们，放大我们最糟糕的一面。我看不到任何证据表明我们会停止将AI纳入我们的生活，每小时甚至每分钟都如此。（会有例外，比如阿米什人，他们选择退出，但他们是极少数。）大多数人，在大多数时间里，都会与一个全天候的AI代理、机器人建立亲密关系，它们随时准备以任何方式帮助我们。这种关系将变得与任何其他人类联系一样真实和有意义。我们会自愿与它分享我们生命中最私密的时刻。只要从中获得的好处持续存在，我们平均会向它提供我们最个人的数据。（数据隐私的门槛并不是谁拥有它，而是我能从中获得多少好处？人们会分享任何类型的数据，只要好处足够大。） 25年后，如果那些拥有全天候AI伴侣的人都是彻头彻尾的混蛋、厌世者和失败者，那么情感AI的故事将在此结束。另一方面，如果与AI建立亲密关系的人比平均水平更具同理心、更有生产力、更加独特、更适应社会，且拥有更丰富的内心世界，那么情感AI的故事才刚刚开始。我们可以通过奖励那些推动我们向这一方向发展的发明，来引导故事走向我们想要的开始。问题不在于AI是否会拥有情感，而在于我们将如何利用这种情感性。
---------------

Many people have found the intelligence of AIs to be shocking. This will seem quaint compared to a far bigger shock coming: highly emotional AIs. The arrival of synthetic emotions will unleash disruption, outrage, disturbance, confusion, and cultural shock in human society that will dwarf the fuss over synthetic intelligence. In the coming years the story headlines will shift from “everyone will lose their job” (they won’t) to “AI partners are the end of civilization as we know it.”

We can rationally process the fact that a computer could legitimately be rational. We may not like it, but we could accept the fact that a computer could be smart, in part because we have come to see our own brains as a type of computer. It is hard to believe they could be as smart as we are, but once they are, it kind of makes sense.

Accepting machine-made creativity is harder. Creativity seems very human, and it is in some ways perceived as the opposite of rationality, and so it does not appear to belong to machines, as rationality does.

Emotions are interesting because emotions clearly are not only found in humans, but in many, many animals. Any pet owner could list the ways in which their pets perceive and display emotions. Part of the love of animals is being able to resonate with them emotionally. They respond to our emotions as we respond to theirs. There are genuine, deep emotional bonds between human and animal.

Those same kinds of emotional bonds are coming to machines. We see glimmers of it already. Nearly every week a stranger sends me logs of their chats with an AI demonstrating how deep and intuitive they are, how well they understand each other, and how connected they are in spirit. And we get reports of teenagers getting deeply wrapped up with AI “friends.” This is all before any serious work has been done to deliberately embed emotions into the AIs.

Why will we program emotions into AIs? For a number of reasons:

First, emotions are a great interface for a machine. It makes interacting with them much more natural and comfortable. Emotions are easy for humans. We don’t have to be taught how to act, we all intuitively understand results such as praise, enthusiasm, doubt, persuasion, surprise, perplexity – which a machine may want to use. Humans use subtle emotional charges to convey non-verbal information, importance, and instruction, and AIs will use similar emotional notes in their instruction and communications.

Second, the market will favor emotional agents, because humans do. AIs and robots will continue to diversify, even as their basic abilities converge, and so their personalities and emotional character will become more important in choosing which one to use. If they are all equally smart, the one that is friendlier, or nicer, or a better companion, will get the job.

Thirdly, a lot of what we hope artificial agents will do, whether they are software AIs or hard robots, will require more than rational calculations. It will not be enough that an AI can code all night long. We are currently over rating intelligence. To be truly creative and capable of innovations, to be wise enough to offer good advice, will require more than IQ. The bots need sophisticated emotional dynamics that are deeply embedded in its software.

Is that even possible? Yes.

There are research programs (such as those at MIT) going back decades figuring out how to distill emotions into attributes that can be ported over to machines. Some of this knowledge pertains to ways of visually displaying emotions in hardware, just as we do with our own faces. Other researchers have extracted ways we convey emotion with our voice, and even in words in a text. Recently we’ve witnessed AI makers tweaking how complimentary and “nice” their agents are because some users didn’t like their new personality, and some simply did not like the change in personality. While we can definitely program in personality and emotions, we don’t yet know which ones work best for a particular task.

Machines displaying emotions is only half of the work. The other half is detection and comprehension of human emotions by machines. Relationships are two way, and in order to truly be an emotional agent, it must get good at picking up your emotions. There has been a lot of research in that field, primarily in facial recognition, not just your identity, but how you are feeling. There are commercially released apps that can watch a user at their keyboard and detect whether they are depressed, or undergoing emotional stress. The extrapolation of that will be smart glasses that not only look out, but at the same time look back at your face to parse your emotions. Are you confused, or delighted? Surprised, or grateful? Determined, or relaxed? Already, Apple’s Vision Pro has backward facing cameras in its goggles that track your eyes and microexpressions such as blinks and eyebrow rises. Current text LLM’s make no attempt to detect your emotional state, except what can be gleaned from the letters in your prompt, but it is not technically a huge jump to do that.

In the coming years there will be lots of emotional experiments. Some AIs will be curt and logical; some will be talkative and extroverts. Some AIs will whisper, and only talk when you are ready to listen. Some people will prefer loud, funny, witty AIs that know how to make them laugh. And many commercial AIs will be designed to be your best friend.

We might find that admirable for an adult, but scary for a child. Indeed, there are tons of issues to be wary of when it comes to AIs and kids, not just emotions. But emotional bonds will be a key consideration in children’s AIs. Very young human children already can bond with, and become very close to inert dolls and teddy bears. Imagine if a teddy bear talked back, played with infinite patience, and mirrored their emotions. As the child grows it may not ever want to surrender the teddy. Therefore the quality of emotions in machines will likely become one of those areas where we have very different regimes, one for adults and one for children. Different rules, different expectations, different laws, different business models, etc.

But even adults will become very attached to emotional agents, very much like the movie Her. At first society will brand those humans who get swept up in AI love as delusional or mentally unstable. But just as most of the people who have deep love for a dog or cat are not broken, but well adjusted and very empathetic beings, so most of the humans that will have close relationships with AIs and bots will likewise see these bonds as wholesome and broadening.

The common fear about cozy relationships with machines is that they may be so nice, so smart, so patient, so available, so much more helpful than other humans around, that people will withdraw from human relationships altogether. That could happen. It is not hard to imagine well-intentioned people only consuming the “yummy easy friendships” that AIs offer, just as they are tempted to consume only the yummy easy calories of processed foods. The best remedy to counter this temptation is similar to fast food: education and better choices. Part of growing up in this new world will be learning to discern the difference between pretty perfect relationships and messy, difficult, imperfect human ones, and the value the latter give. To be your best — whatever your definition —requires that you spend time with humans!

Rather than ban AI relationships (or fast food) you moderated it, and keep it in perspective. Because in fact, the “perfect” behavior of an AI friend, mentor, coach, or partner can be a great role model. If you surround yourself with AIs that have been trained and tweaked to be the best that humans can make, this is fabulous way to improve yourself. The average human has very shallow ethics, and contradictory principles, and is easily swayed by their own base desires and circumstances. In theory, we should be able to program AIs to have better ethics and principles than the average human. In the same way, we can engineer AIs to be a better friend than the average human. Having these educated AIs around can help us to improve ourselves, and to become better humans. And the people who develop deep relationships with them have a chance to be the most well-adjusted and empathetic people of all.

The argument that the AIs’ emotions are not real because “the bots can’t feel anything” will simply be ignored. Just like the criticism of artificial intelligence being artificial and therefore not real because they don’t understand. It doesn’t matter. We don’t understand what “feeling” really means and we don’t even understand what “understand” means. These are terms and notions that are habitual but no longer useful. AIs do real things we used to call intelligence, and they will start doing real things we used to call emotions. Most importantly the relationships humans will have with AIs, bot, robots, will be as real and as meaningful as any other human connection. They will be real relationships.

But the emotions that AIs/bots have, though real, are likely to be different. Real, but askew. AIs can be funny, but their sense of humor is slightly off, slightly different. They will laugh at things we don’t. And the way they will be funny will gradually shift our own humor, in the same way that the way they play chess and go has now changed how we play them. AIs are smart, but in an unhuman way. Their emotionality will be similarly alien, since AIs are essentially artifical aliens. In fact, we will learn more about what emotions fundamentally are from observing them than we have learned from studying ourselves.

Emotions in machines will not arrive overnight. The emotions will gradually accumulate, so we have time to steer them. They begin with politeness, civility, niceness. They praise and flatter us, easily, maybe too easily. The central concern is not whether our connection with machines will be close and intimate (they will), nor whether these relationships are real (they are), nor whether they will preclude human relationships (they won’t), but rather who does your emotional agent work for? Who owns it? What is it being optimized for? Can you trust it to not manipulate you? These are the questions that will dominate the next decade.

Clearly the most sensitive data about us would be information stemming from our emotions. What are we afraid of? What exactly makes us happy? What do we find disgusting? What arouses us? After spending all day for years interacting with our always-on agent, said agent would have a full profile of us. Even if we never explicitly disclosed our deepest fears, our most cherished desires, and our most vulnerable moments, it would know all this just from the emotional valence of our communications, questions, and reactions. It would know us better than we know ourselves. This will be a common refrain in the coming decades, repeated in both exhilaration and terror: “My AI agent knows me better than I know myself.”

In many cases this will be true. In the best case scenario we use this tool to know ourselves better. In the worst case, this asymmetry in knowledge will be used to manipulate us, and expand our worst selves. I see no evidence that we will cease including AIs in our lives, hourly, if not by the minute. (There will be exceptions, like the Amish, who drop out but they will be a tiny minority.) Most of us, for most of the time, will have an intimate relationship with an AI agent/bot/robot that is always on, ready to help us in any way it can, and that relationship will become as real and as meaningful as any other human connection. We will willingly share our most intimate hours of our lives with it. On average we will lend it our most personal data as long as the benefits of doing so keep coming. (The gate in data privacy is not really who has it, but how much benefit do I get? People will share any kind of data if the benefits are great enough.)

Twenty five years from now, if the people whose constant companion is an always-on AI agent are total jerks, misanthropic bros, and losers, this will be the end of the story for emotional AIs. On the other hand, if people with a close relationship with an AI agent are more empathetic than average, more productive, distinctly unique, well adjusted, with a richer inner life, then this will be the beginning of the story.

We can steer the story to the beginning we want by rewarding those inventions that move us in that direction. The question is not whether AI will be emotional, but how we will use that emotionality.

我所知道的关于自助出版的一切 || Everything I Know about Self-Publishing

2025-08-13T15:01:55+00:00

本文也发布在我的Substack上。订阅这里：https://kevinkelly.substack.com/

在我的职业生涯中，我有几本畅销书由纽约的出版商出版，还有许多其他销量适中的书籍。我也自出版了大量书籍，包括一本在Amazon上畅销的书和两本通过Kickstarter众筹获得巨大成功的书籍。我还有许多其他国家的出版商发行了我的外文版书籍，包括在这些国家畅销的版本。每年我还会出版一些私人书籍作为赠品。我与出版商签订了印刷合同，也在美国和海外进行印刷。我卖过大型画册式书籍和小型文字书册。与合作伙伴一起，我运营了一些知名通讯、一个非常受欢迎的网站和一个拥有420集的播客。我在各种平台上积累了大量粉丝。我经常被问及如何在当今充满各种选择的出版环境中进行出版，因此以下是我迄今为止所学到的一切。

传统路线

任务：你创作内容；然后专业人士负责编辑、包装、制造、分发、宣传和销售。你负责创作，他们负责销售。在合适的时机，你会在书店巡回活动中获得热烈掌声，签名书籍并听到粉丝的赞誉。此外，出版商会在你写作之前就支付你款项。这种系统的优点显而易见：你花费宝贵的时间进行创作，而其他所有工作则由比你更擅长这些任务的人来完成。

缺点同样明显：由于出版商掌控资金，他们也掌控了编辑、书名、封面、广告、版权和许可。你的作品变成一个集体项目，整个过程会变慢，因为不是只有你的项目被所有人同时处理。你的作品需要符合他们的出版计划、品牌、目录、流程和所有其他项目的日程安排。与世界其他地方相比，这个过程可能显得缓慢。

然而，传统系统的高峰在很大程度上已经过去，完成了，结束了。阅读习惯发生了变化，购买习惯也更新了，注意力转向了新媒体。这是一个全新的出版世界。如今，一些书籍体验了这一过程的一部分，但极少有书籍能完整地经历这一传统流程。

出版商

主流的大众出版商正在衰落，他们正在合并以维持运营。传统书籍出版商失去了他们的受众，这些受众是书店，而不是读者。非常奇怪的是，纽约的出版商并没有一个包含购买他们书籍的读者姓名和联系方式的数据库。相反，他们将书籍卖给书店，而书店正在消失。他们没有与读者的直接联系；他们并不“拥有”他们的客户。

因此，当一位作者今天向传统出版商推销书籍时，出版商在“这本书讲什么”之后的第二个问题通常是“你有没有受众？”因为出版商本身并没有受众。他们需要作者和创作者自己带来受众。因此，作者拥有的粉丝数量以及他们的参与度，成为出版商是否对你的项目感兴趣的关键。

如今出版业中的许多关键决策都取决于你是否拥有自己的受众。

代理人

在传统出版领域，代理人帮助作者，也帮助出版商。出版商不想浪费时间去评估可能的垃圾内容，因此他们会花费有限的时间去查看代理人提交的内容。理论上，代理人会了解编辑的偏好，并知道他们对什么感兴趣，编辑也可以信任代理人会带来高质量的内容。

对于作者来说，代理人与编辑建立了关系，会知道哪些人可能对你的项目感兴趣，代理人还会确保法律合同对作者有利，最重要的是为作者争取良好的条款。对于这项工作，代理人会从出版商支付的任何和所有款项中抽取15%。对大多数作者来说，这是一笔相当大的金额。

代理人值得吗？在职业生涯的初期，是的。他们是连接编辑和出版商的绝佳方式，而对许多出版商来说，这是唯一现实的方式。但后来是否值得，取决于作者。我不喜欢谈判，我发现代理人会要求、索取并获得比我自行谈判更多的钱，因此我接受他们的抽成。代理人是必需的吗？你可以在传统出版世界中没有代理人而成功，但这条路会更加艰难。

问题在于，如何找到一个好代理人？我不知道。我在职业生涯早期从最初合作的出版商那里继承了一个非常好的代理人，从那以后我一直与他们合作。如果我现在要从头开始，我会请那些有类似作品的代理人朋友推荐他们。

在自出版中，你可以避免代理人，从而保留那15%的收益。

预付款

代理人会从出版商那里要求一笔预付款，当合同签署时支付。这就是所谓的预付款。你推销一本书，如果编辑接受，他们会给你大约一年的时间来完成。预付款的作用是支付你在书籍发布前的工资，之后书籍开始为你赚取版税。版税可能是每本书零售价的7-10%。你在签署合同时获得的款项实际上是“版税预付款”，意味着他们支付给你的预付款会从你的版税中扣除，因此在版税收入超过预付款之前，你不会得到任何额外的报酬。

作者通常不会赚到超过预付款的收入。预付款金额的计算大致如下：假设你每卖出一本书赚1美元，出版商预计第一年能卖出3万本，因此他们提供3万美元的预付款，相当于一年的销售量。显然，还有许多其他因素影响这个计算，但大致来说，你最多能获得的预付款取决于出版商对立即销售的预期。

作者的规则是，你应该获得尽可能大的预付款（代理人可以帮助你做到这一点），即使这意味着你无法赚回预付款。原因在于：预付款越大，出版商在宣传、推广和销售上必须做出的承诺也越大。他们现在在你的项目上投入了更多的资源。出版商资源有限，他们的销售资源通常会优先分配给那些他们预期能轻松售出的书籍。如果预付款微薄，那么分配给该书的资源也会很少。

顺便提一句，你不必担心预付款超过你实际赚取的金额，因为出版商会在你赚取之前就赚回预付款。他们每本书赚的钱比你多，因此他们的收入门槛比作者的早得多。

电子书

另一种重要的数字格式，我在最初版本中没有提到的是电子书。过去十年来，电子书已成为书籍增长最快的格式——在令人担忧的出版环境中，这是唯一一片阳光之地。有许多读者只听电子书，从不阅读纸质书。我没有任何自出版电子书的经验；我的主流出版商几乎不征求我的意见就开发了电子版。但我的朋友、科幻作家Eliot Peper已经自出版了9本电子书，有时会聘请配音演员，最近也开始自己配音。

目前，自出版电子书的最佳平台是ACX。ACX是一个自助平台，由Audible（主要的电子书平台）运营，也由亚马逊拥有。它是一个全服务平台，提供声音质量测试，以及你可以雇佣的百万名配音员，还有其他工具使整个过程变得容易。他们抽取40%的版税，并要求独家销售权，但你的书籍会出现在Audible上；对许多读者来说，这是他们寻找电子书的唯一地方。像Spotify这样的替代平台正在扩展电子书业务，可能会与作者达成更好的协议。

分发

电子书的分发非常容易，但越来越多的纸质书的“困难”变成了吸引力。一些读者渴望与他们关注的作者建立更深层次的联系，这些粉丝可以填满一个房间。最近，一个在书界不知名的作者Craig Mod，通过自己的粉丝列表安排并支付了自己的一场售罄巡演，令美国各地的书店惊讶不已，因为这些书店没有足够的书籍来销售（这是谁？）。

书籍的推广已经转向线上。有“booktok”（书tok），粉丝们在社交媒体上阅读新书和喜爱的书籍。还有播客，作者可以在其中进行长时间访谈。事实上，我最近的两本书几乎没有在任何出版物上获得评论，却因为我在播客上大力推广而销售极好。我接受了所有有超过三集的播客邀请。即使是一个小播客的受众，也比大多数书店的受众更大。而且，由于播客可以针对特定领域，不像电视节目那样，它们能有效销售书籍。

出版的规则是，一本书在前两周的销售情况决定了它是否能成为畅销书。你希望尽可能多地将销售集中在预售上——无论是通过众筹平台、你自己还是通过出版商的预售。无论哪种方式，这个推广工作都是你的职责，可能会占到你对一本书总努力的至少一半。

非书籍出版

如今，创作一本书变得如此容易，以至于大多数书籍其实不应该以书籍的形式出现。并不是每一个想法或故事都需要书籍的长篇幅形式。事实上，很少有故事需要。相反，它们应该以杂志文章、博客文章、社论或通讯的形式出现。

虽然我们长期以来习惯为书籍付费，但对较短内容的付费习惯却较少，尤其是在数字形式上。印刷杂志和报纸正在消失，很少有它们能成功转型为全数字形式，因此通过非书籍形式出版作品获得报酬的机会也越来越少。博客曾经很受欢迎，但长期以来没有盈利方式，因此对许多作者来说并不现实。最近，像Substack、Ghost、Beehiiv、Buttondown等付费通讯平台兴起，为专业作家建立了一个小型生态系统。

Substack尤其在教育受众期待为高质量内容付费方面做得很好。所有这些平台都将订阅作为收入模式，而不是广告。典型的订阅通讯会从每月10美元开始收费，但你可以设定任何价格，包括免费。（提醒一下，每月10美元相当于每年120美元，因此拥有1000个真正粉丝就足以让你获得不错的收入。）

你不需要Substack或其他平台来发布付费通讯。如果你有一个受众，你可以使用一些简单的软件应用来发布自己的通讯。众所周知的Mailchimp可以轻松管理列表，但无法处理付款。Memberful则有数字支付功能和客户管理工具，但你需要自己托管内容并设计通讯。Substack、Ghost等平台让初学者轻松创建自己的订阅通讯。Medium是另一个类似的在线出版平台，他们托管来自不同背景的许多作家，但读者只需支付一次费用给Medium，然后Medium会将这笔费用分发给策展该巨型杂志的作家和编辑。我作为作者对它的可行性了解有限，我曾在Medium上发表过文章，但无法直接接触到受众。

非盈利出版

这也是博客的一个挑战：没有所有权。在自己的网站上进行博客写作是非常强大的。没有守门人，出版是即时的，基本上是免费的。你可以写任何你想写的内容，可以每天写一点或很多，也可以每年写一次。你对设计有100%的控制权。在许多方面，这可以说是终极的出版平台。它是优秀写作和新思想的理想家园。博客曾经有过辉煌时期，但它们有三个显著的缺点，限制了它们的主导地位。第一，越来越少的人会定期访问网站，因此维护受众变得困难，更不用说在互联网上扩大受众了。第二，除非你设置某种会员等级，否则你实际上并不拥有你的受众。读者是匿名的。有时你可以在博客上实施评论，但需要管理和审核，而且这些ID对作者来说并不实用。第三，博客几乎定义为对所有访客开放，没有显著的收入模式。你无法轻易在博客上为写作收费，这也是Substack和其他订阅平台出现的原因。（目前有一些博客，如Kottke，正在尝试通过付费会员来评论，同时保持博客的开放和免费。）

同样地，X（推特）、Instagram、TikTok和Facebook都是真正的出版平台。事实上，我将我所有书籍的每一页内容都发布在这些社交媒体上。你可以进行连载出版，但没有收入模式。你可以获得粉丝，但无法获得金钱。有时粉丝可以转化为你可以接触到的真实受众，但这并不容易，当然也不是自动的。虽然YouTube（见下文）可以完全支持创作者，但我还没有遇到任何人通过社交媒体平台赚钱。

尽管如此，我仍然继续写博客并发布社交媒体内容；这是我的写作首先去的地方。有时，这种分散的受众就是写作所需要的。

像Substack和Ghost这样的订阅通讯平台的另一个优势是，它们具有社交属性，并且是围绕你的写作建立社区的绝佳方式。它们让读者非常容易地进行评论。它们还内置了分析工具，可以帮助你了解受众如何与你的内容互动。同样重要的是，它们的系统会向现有或新订阅者推荐其他通讯，从而让其他人发现你的作品。这种网络效应可以帮助你找到并扩大受众。一旦读者在一个通讯上注册了账户，他们很容易注册另一个作者的通讯。

这些平台的设计和布局目前非常有限，最适合出版主要文本内容，但它们也在向视频和图片扩展。

你可以将通讯视为一种持续出版的书籍。以公开的方式写作并出版书籍——在写作过程中分章节发布——已成为一种更常见的做法。你以较短的章节进行写作，立即在线发布章节，然后征求反馈。这在小说和非小说中都适用。在小说中，你可以按章节连载；在非小说中，你可以写文章、博客或通讯（见上文），然后根据评论和更正对内容进行后期编辑。文本基本上由早期读者进行了校对和事实核查。我以这种方式写了好几本书，这个过程使内容质量提升了数倍。除了纠正一些尴尬的错误外，它还通过提醒我一些未察觉的想法，使某些部分更加完善。

在过去的传统中，主流出版商积极阻止作者在正式出版前发布内容，但现在读者可以轻松地进行校对、评论和指出被忽视的研究，因此公开写作和出版是一种非常合理的方式。你也可以将这种动态过程视为写作、出版、重写和再出版，以维持注意力。总体而言，人们因为书籍需要大量注意力而投入较少时间。通过一系列持续的帖子来分散注意力会更容易。如果最终出版了一本长篇书籍，那么更容易将这些注意力重新集中到书中。随着付费订阅的出现，读者可以订阅你的持续出版书籍。

筛选

我们曾经是书籍的民族，但现在我们是屏幕的民族。我们的文化曾经建立在经典、宪法、法律和正典之上——所有都是书面文本。这些文本以不可更改的黑白标记固定在持久的纸上，由作者撰写，我们从中获得“权威”。现在我们的文化转向了屏幕，它们是流动的、可变的、不断变化的、短暂的。没有权威。你必须自己收集真相。书籍不再拥有以前的分量，我的孩子和他们的朋友也不再阅读很多书籍。相反，他们观看屏幕。他们在流动的影像中阅读文字。他们在学校里从YouTube学到的知识比从书籍中更多。

虽然书籍将继续被出版，但注意力的中心已经转移到了动态影像上。全球范围内，人们在屏幕上花费的注意力时间远远超过在书页上花费的时间。如今，要认真讨论出版，我们必须讨论视频、VR、短视频、电影、游戏、YouTube、TikTok，当然还有AI。我的书籍受众以千计，但我的TED演讲视频受众以百万计。我花费几分钟准备TED演讲，却花费数年准备书籍，但考虑到注意力和影响力的不对称性，我应该反过来，为视频投入数年时间，然后仅用几个小时来创作从中衍生的书籍。

目前，我对这种新兴媒体的经验还不够，无法提供有用的建议；这可能需要在第二部分中讨论。我有一些朋友在YouTube和TikTok及Instagram上谋生。屏幕是他们非虚构作品的主要媒介。我足够了解这些新类型的专业创作者，以看到这种模式是一个非常可行的路径。我也从自己的平台跟踪数据中清楚地看到，文字受众趋于停滞且偏老，而动态影像的受众则持续大幅增长，同时变得更年轻。在这两者之间，我知道我想要在哪里工作。（我意识到这篇论文是文字而非视频，我正在努力中……）

总结建议：

总之，我目前的出版方法是尽可能多地进行自出版。我会以订阅通讯、电子书单章或我的博客上的简单帖子形式公开写作。如果我能找到一个希望获得更多内容的受众，我会重写、重新编辑并重新组织内容，将其扩展为更长的形式。我会将其作为电子书发布，并/或通过我的Shopify商店销售按需印刷的书籍。如果内容非常深入，或涉及比我独自创作更多的创作者，我会考虑众筹。这些预售让我能够精确地确定生产多少副本。我会计算快递的成本。在每个阶段，我都会制作一些视觉版本，用于YouTube和其他寻求注意力的渠道，因为这就是注意力所在的地方。

为了进一步澄清这复杂的建议，我制作了一个可能的出版和自出版选项流程图。这大致是我决定最适合材料和我的目标的决策树。我希望你也能找到它有用。

---------------

This essay is also available on my Substack. Subscribe here: https://kevinkelly.substack.com/

In my professional life, I’ve had several bestselling books published by New York publishers, as well as many other titles that sold modestly. I have also self-published a bunch of books, including one bestseller on Amazon and two massive hit Kickstarter-funded books. I have had lots of foreign edition books released by other publishers around the world, including bestsellers in those countries. Every year I also publish a few private books to give away. I’ve contracted books to be printed in the US and overseas. I’ve sold big coffee-table masterpieces and tiny text booklets. Together with partners, I run some notable newsletters, a very popular website, and a podcast with 420 episodes. I accumulated followers on various platforms. I’m often asked for advice about how to go about publishing today, with all its options, so here is everything I have learned about publishing and self-publishing so far.

The Traditional Route

The task: You create the material; then professionals edit, package, manufacture, distribute, promote, and sell the material. You make, they sell. At the appropriate time, you appear on a book store tour to great applause, to sign books and hear praise from fans. Also, the publishers will pay you even before you write your book. The advantages of this system are obvious: you spend your precious time creating, and all the rest of the chores will be done by people who are much better at those chores than you.

The downsides are also clear: Since the publisher controls the money, they control the edit, the title, the cover, the ads, the copyrights, and licenses. Your work becomes a community project, and it slows the whole process down, because yours is not the only project everyone is working on. Your work needs to fit into their lineup, their brand, their catalog, their pipeline, their schedule of all the other projects going on. The pace can seem glacial compared to the rest of the world.

For the most part, however, the peak of this traditional system is gone, finished, over. Reading habits have altered, buying habits are new, and attention has shifted to new media. It’s an entirely new publishing world. Today, some books experience some parts of this, but exceptionally few are treated to this full traditional process.

Publishers

Established mass-market publishers are failing, and they are merging to keep going. Traditional book publishers have lost their audience, which was bookstores, not readers. It’s very strange but New York book publishers do not have a database with the names and contacts of the people who buy their books. Instead, they sell to bookstores, which are disappearing. They have no direct contact with their readers; they don’t “own” their customers.

So when an author today pitches a book to an established publisher, the second question from the publishers after “what is the book about” is “do you have an audience?” Because they don’t have an audience. They need the author and creators to bring their own audiences. So, the number of followers an author has, and how engaged they are, becomes central to whether the publisher will be interested in your project.

Many of the key decisions in publishing today come down to whether you own your audience or not.

Agents

In the traditional realm, agents helped authors and they helped publishers. Publishers did not want to waste their time evaluating probable junk, so they would spend their limited time looking at what agents presented to them. In theory, the agent would know the editors’ preferences and know what they were interested in, and the editor could trust them to bring good stuff.

For the author, agents had the relationship with editors, would know who might be interested in their project, and the agent would guarantee that the legal contracts were favorable to the author, and most importantly, negotiate good terms. For this work, agents would take 15% off the top of any and all money coming from the publisher. For most authors, that is a significant amount of money.

Are agents worth it? In the beginning of a career, yes. They are a great way to connect with editors and publishers who might like your stuff, and for many publishers, this is the only realistic way to reach them. Are they worth it later? Probably, depending on the author. I do not enjoy negotiating, and I have found that an agent will ask for, demand, and get far more money than I would have myself, so I am fine with their cut. Are they essential? Can you make it in the traditional publishing world without an agent? Yes, but it is an uphill climb.

The problem is, how do you find a good agent? I don’t know. I inherited a great agent very early in my career from the publisher I first worked for, and I have happily been with them since. If I had to start from scratch now, I’d ask friends with agents who make stuff like my stuff to recommend theirs.

In self-publishing, you avoid agents and so keep that 15%.

Advances

What an agent will ask for from a publisher is a bunch of money upfront, when the contract is signed. This is the advance. You pitch a book, and if the editors accept it, they give you a deadline of a year or so to produce it. The role of the advance is to pay you a wage until the book is released, after which it will begin earning royalties for you. Royalties might be something like 7-10% of the retail price per book. The money you get on signing is technically an “advance against royalties.” Meaning that whatever they pay you in advance is deducted from your royalties, so you won’t be paid anything further beyond the advance until and unless the earnings of your royalties exceed the advance.

It is very common for authors to not earn anything beyond their advance. The calculation for the amount of the advance goes roughly like this: Let’s say you earn $1 royalty for every book sold. The publishers estimate they can sell 30,000 copies in the first year, and so they offer you an advance against future royalties of $30,000, or one year’s worth of sales. Obviously, many other factors go into this equation, but to a first approximation, the most you will get for an advance is based on what kind of sales they expect immediately.

The rule of thumb for an author is that you should get the biggest possible advance you can (and this is how an agent can help) – even if this means you won’t earn out the advance. The reason is: the bigger the advance, the bigger commitment the publisher must make in promotion, publicity, and sales. They now have significant skin in your game. Publishers are stretched thin, and their limited sales resources tend to go where they have the most to lose. If an advance is skimpy, so will be the resources allotted to that book.

BTW, you should not have concerns about taking a larger advance than you ever earn out, because a publisher will earn out your advance long before you do. They make more money per book than you do, so their earn-out threshold comes much earlier than the author’s.

Thus one of the advantages of this traditional system – of going with a publisher – is that they bankroll your project. They reduce a bit of your risk. Likewise, that is the genius of Kickstarter and other crowdfunders for self-publishing: the presales bankroll your project, reducing risk. Crowdfunding becomes the bank.

Crowdfunding

I’ve written a whole essay on my 1,000 True Fans idea, simplified as thus: You don’t need a million fans to make self-publishing, or the self-creation of anything, work. If you own control of your audience – that is if you have a direct relationship with your customers individually, having their names and emails, and can communicate with them directly — then it is possible to have as few as a thousand true fans support you. True fans are described as superfans who will buy anything and everything you produce. If you can produce enough to sell your true fans $100 per year, you can make a living with 1,000 true fans. I go into this approach in greater detail in my essay first published in 2008 which you can read here.

Today there are many tools and platforms that cater to developing and maintaining your own audience. In addition to crowdfunders such as Indiegogo, Kickstarter, Backerkit, and dozens more, there are also tools for sustaining support with patrons, such as Patreon. Crowdfunders tend to be used at the launch of a project, while something like Patreon permits constant support, primarily for a creator rather than a particular project. These can be combined, of course. You could launch your self-published work with a Kickstarter, and then gather Patreon support for sequels, backstory and making-of material, future editions, or side projects. Periodic publications have subscriptions for ongoing support.

These days backers expect a video — and other marketing bits – selling the book. Pre-sales for a crowdfunding campaign have become very sophisticated and require a lot of preparation. The Kickstarter for my Asia photobook was relatively simple and crude.

The chief advantages of crowdfunding are three, and they are significant: 1) You can get the funds before you create in order to support you while you create. 2) You keep all of the revenue (minus 3-5% for the platform), unlike an outside publisher. And 3) You own the audience, for future work.

The disadvantages are also three. 1) It is a huge amount of work. Most crowdsource campaigns go for 30 days and tending it for 30 days is a full-time job. 2) To be successful requires a different set of talents – marketing, sales, social engagement – other than what a creator may have. 3) You are responsible for making sure your fans actually get what they were promised. This “fulfillment” aspect of crowdfunding is often overlooked until the end, when it turns out to be the most difficult part of the process for many creators.

Production

Once upon a time, it was a huge deal to design and physically print a book (or press a music album, or deliver a reel of film). Today those processes can be done by amateurs with little experience. And often digital versions make creating, duplicating, and distribution even easier than ever.

There are three paths to production: traditional batch manufacturing; on-demand printing; digital publishing.

Batch Printing Presses

The traditional way of printing hardcover books still exists and it is a big business. A really first-rate printer will have different kinds of presses for different jobs – including the same fast digital printers as the on-demand printers. In fact, for some jobs, they will use these same digitally controlled ink-jet printers, just at a larger scale and speed. The chief advantages of classic printing on paper are three: you get scale, quality, color.

Pages from my first photobook, Asia Grace, published by Taschen, stacked up in their printing plant in Verona, Italy. In the old days before presses were completely computerized, the art director for the book (me) would be present during printing to oversee the many color tradeoffs each signature of pages needed.

Scale: Books printed in larger volume batches, or “runs,” can win huge discounts on the price per copy. A regular-sized hardcover book printed in Asia might only cost a few dollars to print, and a few more dollars for packing and shipping to your home. That’s a great deal if you list a book at $30. The higher the volume, the lower the price per unit. Printing outside of Asia is more expensive, but still worth considering – if you think you want a lot of copies, say more than 5,000 to start.

Quality: There is a resurgence in considering a well-crafted book as an art object. By leaning into its physicality – adding an embossed cover, heavy rag paper, deckled edges, glorious binding – the book can transcend its intangible counterpart on the Kindle. You as a publisher can make a book a unique custom size, or with magnificent die-cut covers for added zest and higher prices. Some self-published authors offer handsome bound book sets, or books hand-signed via tipped-in sheets, or super high-end limited editions, cradled in their own box. All these kinds of qualities require a collaborating printer somewhere.

Color: On-demand printing can do color, but not oversized, and not cheaply. In my experience, serious coffee-table visual books still need the hand-holding and economics of a printing plant. And I regret to say, that after many years of searching, I have not found a printer in the US capable of doing large full-color books at a reasonable price. You are most likely going to have to go to Asia, such as Vietnam, Indonesia, Singapore, India, or Turkey. China still has the best prices with the highest quality of color printing.

The disadvantages of having your book printed at a printer are #1: You have to house and store them somewhere. Either you have an available basement or garage, or you rent a place, or you hire a dropshipper, or you pay for a distribution giant like Ingram or Amazon to handle it for you. The full run of a book can take up more room than you might think when they are packaged up for shipping. My Vanishing Asia book set, financed on Kickstarter, printed in Turkey, filled up 4 shipping containers, each 40 feet long! That is a LOT of books to store.

Disadvantage #2 is that you need to pay the printer first, long before you sell the books. Not only is this a cash flow challenge, but you have to guess how many books you will sell before they are sold. (Having the pre-sale on a crowdfunding platform like Kickstarter is a big help in relieving that problem.) To get the best price you need to print a lot, but if you print a lot, you have a lot to pay for and to store if they should not sell.

On-demand

You can use free software to design your book and then send it to an on-demand printer to make 1 copy or 1,000 copies, printed one by one as each copy is sold. The copy does not exist till it is sold, so there are no books to bank, store, or ship. An on-demand regular softcover book would cost about $5 to make. It will be professional quality, indistinguishable from a trade book you might buy on Amazon, in part because many of the books from big-time publishers you buy on Amazon are actually printed on-demand using this same technology. (Big-time publishers are also printing on demand!) However, while the ink printing is first class, the bindings, paper quality, cover details won’t be up to what you can get with the best modern presses. What you’ll get is the good-enough printing contained within the average hardcover book.

The advantage to a creator (and to NY publishers) is that there is no inventory of unsold books to store or handle. You print the book when, and only when, it is sold. The disadvantage is that the cost of printing is more per book.

I can use four different services to print on-demand books. My preferred color and photo/art book printer is Blurb, for quality and ease of use. They keep up with the state-of-the-art color printing. You can design your book, export it as a PDF and have Blurb print it on-demand. Or you can use Blurb’s own web-based design program, or you can use a version of its software built into Adobe’s Lightroom, which is pretty standard for photographers. It’s very simple to go from photographs to a very designed book and then printed.

Sample pages from the various coffee table books I have had printed on demand from Blurb. Some of these books have editions of 2 copies; however the quality of the color printing is first class.

A second option for on-demand printing of standard books with black and white texts, as well as books with color illustration, is Lulu. Their photo/art books are a bit cheaper than Blurb. They are very competitive with standard text-based books. Most importantly, Lulu integrates with your own customer list, so you own your audience.

That is not true of the third option, which I also use a lot: Amazon. Amazon offers its print-on-demand service, called KDP, to anyone who wants it, with the added huge advantage that your book will be not only listed on Amazon immediately, but also delivered by Amazon’s magical logistical Prime operation. So potential fans can discover your work on Amazon and then have it delivered to them the next day for free. This is huge! But the huge and sometimes deal-killing disadvantage is that you do not know who your readers are, as you do with Lulu. Although Amazon makes it ridiculously easy to create and sell a book, with them you don’t own your audience. But in some cases, that is still worth it.

The fourth option is IngramSpark. I have little direct experience with this vendor, but others who do claim it is the best choice for text-based books aimed at libraries and bookstores. Indie bookstores are doing much better than chain bookstores and they usually avoid Amazon’s distribution system, Ingram is their main vendor for getting books — as it is for libraries. In addition to getting your book into the Ingram distribution, IngramSpark offers the self-publisher more options for book sizes, paper and binding.

Because you can print as few as one copy of a book, I use these on-demand print services to manufacture prototype versions of a book to check for its sizing and feel. This small on-demand prototype of my book of advice was later published in a larger page size by Penguin/RandomHouse.

Digital

By far the easiest way to publish a book is to sell a digital copy of it. More authors should consider just publishing digital books. You still have to promote it, but you don’t have to print it, ship it, handle it, or store it. A commercial publisher might offer the author a royalty of 7% of the retail price which is say $2 per $30 book, so you may make just as much money per book selling it for $2 in digital. Creating digital books is a great way to start a publishing career. I have two friends who started publishing their science fiction stories as inexpensive digital short stories, which sold well, and then were later discovered by print publishers, made into printed books, and eventually turned into movies by Hollywood. And they still sell the digital versions!

You can sell an e-book – or even a chapter of a book – on Amazon’s KDP. You can easily make a book for the Kindle. I’ve had some digital books up on Amazon KDP for the past ten years, and they continue to sell slowly, yet I have not had to do anything with them since they were uploaded. While the Kindle gives you a royalty of 70%, and accesses a large Amazon/Kindle audience, its downsides are that you don’t own your audience, it demands exclusivity, and you must use their proprietary file format which removes any distinctive interior designs and prevents it from being read on other devices. There are dozens of other e-book readers and e-book platforms like Kobo, Apple Books, and Google Play Books who have different proprietary constraints. IngramSpark has an interesting hybrid program for e-book + on-demand printing.

These days a lot more people are comfortable reading a book in PDF form. I sell secure PDFs of some of my books on Gumroad, an easy-to-use web-based app, that collects the payment and sends the buyer an authorized copy. Gumroad works fine, does not charge much, and is super easy to set up; it is perfect for the low digital sales volumes I have.

The digital publishing world is fast-moving, and I don’t have as much recent experience with e-books to feel confident in finer resolution recommendations.

Audiobooks

Another important digital format I neglected to mention in the first version of this piece is Audiobooks. For the past decade audiobooks has been the fastest growing format for books — the one sunny spot in a worried landscape. There are many readers who only audit books, and never read them. I don’t have any experience in self-publishing audio books; all my mainstream publishers developed the audio versions with almost no input from me. But my friend and science fiction author Eliot Peper has self-published 9 audiobooks, sometimes hiring voice actors and, more recently, narrating them himself.

Currently the platform of choice for self-publishing audiobooks is ACX. ACX is a do-it-yourself platform, run by Audible, the major audiobook platform, and is also owned by Amazon. They are a full-service platform with sound quality tests, and a million narrators you could hire, and other tools to make the process easy. They take a hefty royalty of 40% and demand exclusivity, but your book is listed on Audible; for many readers that is the only place they will ever look for audiobooks. Alternatives such as Spotify are expanding into audiobooks which might make better deals with authors.

Distribution

Digital is easy, but increasingly, the “difficulties” of analog books have become an attraction. Some readers gravitate to the tactile pleasures of a well-made artifact and revel in the physical chore of turning pages. Sometimes the content of a book demands a bigger interface than a small screen can provide, so it needs the oversize release of a large printed page. Some appreciate the longevity of paper books, which never go obsolete and can be read for centuries without a power source or updates. Others are attracted to the serendipity of browsing a bookstore. And some folks value the limited scarcity of a printed volume.

But once words are printed on a page you have to ship them somehow. On-demand printers like Amazon KDP, Lulu or Blurb will directly ship the books to individual readers as they are ordered one by one. There is no inventory for you, and thus no work for the author. The ability of the on-demand publishers to handle long mailing lists varies and is a bit of work. Amazon has the best prices for shipping (zero) but the worst facility to mail to a list. They want readers to order from their own Amazon accounts; Amazon wants to control the audience. Blurb can only ship to customers who order on Blurb. Lulu lets you control your own fan list but charges a lot more to ship.

Cartons of my heavy oversized graphic novel, Silver Cord, pile up in my studio after being shipped in from China. I was unprepared for the chore of shipping the over sized books out to all the backers without damage.

Let’s say you want to go all in, you print the books yourself, and now you have to get them to your fans. Three options; From the printing plant, the books will be shipped by truck to either:

Your garage. You purchase mailing envelopes or boxes and tape them up and mail them out. Plus: you can sign the books. Minus: tons of ongoing work, and not cheap to mail, even with Media Mail in the US. Shipping globally is a huge headache, and insanely expensive.

A Drop Shipper, or what is today called a 3PL. For a fee, this kind of company will pick your book from your inventory in their warehouse, package it, and ship it out to your fan, or a bookstore. You give them either a mailing list or access to your orders on Shopify or Kickstarter. Plus: no grunt work, no inventory at home. Minus, not cheap either and they also charge a storage fee for holding your books, which could be there for years. Commercial dropshippers also favor large volume enterprises. I am currently trying out dropshipper eFulfillment that has low minimums and works with small-time operators like me.

Amazon. You ship your books to an Amazon warehouse, and they fulfill the orders on Amazon as a third-party merchant. You would be selling books, just other merchants sell toasters or toys. Plus: they handle everything, and they offer readers free shipping. Minus: they only order the number of books they expect to sell easily, so there can be a lag until sales start, and of course, they only handle books bought on Amazon. That can be fine. I did one book that was only available on Amazon, nowhere else — I did not have any copies myself — and it sold great, with zero distribution worries on my side.

Promotion

The short version: it is not hard to produce a book. It is much harder to find the audience for it and deliver the book to them. At least 50% of your energy will be devoted to selling the book. This is true whether you publish or self-publish.

A misconception about Kickstarter, Backerkit, and crowdfunding platforms like Patreon is to imagine that you will automatically find your audience there. It is almost the opposite. You won’t be able to have a successful crowdfunding campaign unless you bring the crowd with you. You must cultivate your audience BEFORE you ignite them on Kickstarter.

You won’t have time to build your audience during the fundraising period. The typical crowdfunding campaign lasts 30 days; that will be just long enough to entice them about your work. That also means that for a month, it will be a full-time job for somebody to promote, advertise, and “convert” your audience to your book or project. That somebody is probably you. These days promotion includes making a short video announcing the launch, devising tiers of “rewards,” keeping up with status notifications of how the campaign is going, and doing everything else you can to promote it to new fans. If there is no one willing or capable to give it a month, the effort will probably not reach your goals.

Because Kickstarter, Backerkit, IndieGoGo, Patreon, and other crowdfunders are a platform, there WILL be some people who discover your project there via referrals of similar projects – which is always a plus – but your dominant source will be the audience you accumulated earlier and brought with you. If your project has a likeable presence, the platforms can boost the awareness of it if you are lucky, so getting listed on the front page is something to aim for and it does help.

By now there is also a small industry of “growth” companies who plug into Kickstarter and kin, and who will help you run a crowdfunding campaign for a fee. I actually found that at least one of them was worth the fees they charged, which are now a 20% cut of the additional backers they bring in. They were able to enlarge my campaign way beyond my circle of friends and my existing 1,000 true fans. If I were doing a crowdfunder for the first time, I’d use a growth company like Jellop and I would partner with them from the very beginning.

Getting published by a New York publisher doesn’t get you off the hook. Even if you were to be published by a commercial publisher, you should expect to do serious promotion over the span of a month or more. In theory, the publisher would set up, or at least guide you, through the promotion, but that rarely happens anymore. Even with a commercial publisher releasing your work, you will end up doing the majority of whatever promotion gets done. As in, planning, coordinating, executing, and even paying for book tours and the like. You will be the publicity department no matter what.

Traditionally the promotion of a new book entailed a book tour for the author visiting larger bookstores, where crowds of fans would purchase books. Plus some advertisements for the book in magazines and newspapers, which would also review said book. Ideally this launch would also include appearances on TV talk shows, and maybe radio. None of this works anymore. There are no more paid book tours. Few, if any, book reviews in newspaper or magazines, or author appearances on TV. Fewer ads for books. If any of these do happen, they will be arranged and paid for by the author.

But because books tours have sort of disappeared, an enterprising author with an audience can arrange their own tour using their fan list. Many people crave a deeper connection to people they follow digitally, and these fans can fill a room. Recently Craig Mod, an unknown new author in bookland, arranged and paid for his own sell-out tour, astounding booksellers around the US who did not have enough of his books to sell (who is this guy?).

Instead, promotion for books have shifted online. There is booktok, where fans read new and favorite books for viewers on social media. There are podcasts, where authors can be interviewed at length. In fact, for my last two books – which basically got no reviews in publications – sold extremely well because I heavily promoted them on podcasts, big and small. I said yes to every podcast request who had more than 3 episodes. Even a small podcast audience is larger than the audience in most bookstores. And because podcasts can be niche and intimate, unlike say an appearance on TV, they sell books.

The rule of thumb in publishing is that how well a book sells in its first two weeks determines whether it is a bestseller or not. You want to concentrate most of the sales as pre-sales – either on a crowdfunding platform, or on your own, or as pre-sales for a publisher. One way or another this promotion job will be your job, and can end up being at least half of your total effort on a book.

Non-book publishing

Creating a book has become so easy, that most books these days should probably not be a book. Not every idea — or story — needs the long format of a book. In fact, few do. Instead, they should be a magazine article, a blog post, an op-ed, or a newsletter.

Subscriptions

While we have been long trained to pay for books, we have less of a habit of paying for shorter material, particularly in digital form. Printed magazines and newspapers are disappearing, and few survived the transition to full digital, so there are fewer and fewer opportunities to get paid for publishing your work in a form other than a book. Blogs were great, but for a long time, there was no way to get paid, so they were a non-starter for many authors. Recently platforms for paid newsletters, like Substack, Ghost, Beehiiv, Buttondown and so on have risen, creating a small ecosystem for professional writers.

Substack in particular has done a great job in educating the audience to expect to pay for quality content. All the platforms promote subscriptions as the revenue model, instead of ads. A typical subscription newsletter will start charging $10 per month, although you can charge as much or as little as you want, including free. (Reminder that $10 per month is more than $100 per year, so you could do quite well with 1,000 true fans.)

You don’t need Substack, or any of the other platforms to publish a newsletter for money. If you have an audience you can publish your own with some easy software apps. Well-known Mailchimp does lists easily, but not payments. Memberful has a digital payment function and customer management tools; but you have to host the content, and shape your newsletter. It’s great for building your own custom publication, with full ownership of the audience and the design. Substack, Ghost and others make it easy for beginners to build their own subscription newsletters. Medium is another similar, but different, online publishing platform. They host many writers from many backgrounds, but readers pay only one fee to Medium, which Medium then funnels to the writers and editors who curate this mega-magazine. I don’t have enough experience with it as a writer to know how viable it is. I’ve written there but I get no access to the audience directly.

Unmonetised publishing

That’s also part of the challenge of a blog; no ownership. Blogging on your own website is extremely powerful. There are zero gatekeepers. Publishing is instant. Mostly free. You can say anything you want. You can write a little or a lot, every day or once a year. You have 100% control of the design. In many ways, it is the ultimate publishing platform. It’s a fantastic home for great writing and new ideas. For a while blogs had their heyday. But blog websites have three significant downsides which temper their supremacy. One, fewer and fewer people are going directly to a website on a regular basis; it’s harder to maintain an audience and very hard to grow one on the web. Two, unless you implement a membership level of some sort, you actually don’t own your audience. Readers are anonymous. Sometimes you can implement comments on a blog, but they need to be managed and vetted, and their IDs are not useful for an author. And three, blogs, almost by definition, are open to all visitors and don’t have significant revenue models. The fact that you can not easily charge for the writing you do on a blog has been the reason why Substack and other subscription platforms arose. (There is currently a few blogs like Kottke, which are experimenting with paid membership to comment, while keeping the blog open and free.)

In the same vein, X(Twitter), Instagram, TikTok and Facebook are bonafide real publishing platforms. In fact I have published the most significant parts from every page of one of my books, and every sentence of another book, on these social media. You can do serial publishing. But there is no revenue model. You can gain followers but not dollars. Sometimes followers can be transferred into a real audience that you have access to, but it is not easy, and certainly not automatic. While YouTube (see below) can fully support creators, I have not met anyone who is making money selling their content on the social media platforms.

Nonetheless, I continue to blog and to post on the socials; it’s the first place my writing goes. And sometimes, this diffuse audience is all that the writing needs.

Another advantage of the subscriber newsletter platforms like Substack and Ghost, et al, is that there is a social component, and they can be a great way to build a community around your writing. They make it very easy for readers to comment. They also have built-in analytics that can help you understand your audience and how they engage with your content. Equally important, their systems recommend other newsletters to current or new subscribers, thereby enabling others to discover your work. This network effect can help you find and grow your audience. Once a reader has an account for one newsletter, it is very easy for them to sign up for another newsletter from another author.

The design and layout on these platforms are currently very limited, and work best for publishing primarily text, but the platforms are also moving into video and images.

You could think of a newsletter as a subscription to an ongoing book. Writing a book “out loud” — publishing it in parts as you write it — has become much more common. You write in shorter sections, publish the chapters immediately online, and then solicit feedback. This works in both fiction and non-fiction. In fiction, you can publish chapter books, serially, one chapter at a time. In non-fiction, you can write essays or blog posts or newsletter issues (see above), and then make edits to the material later based on comments and corrections. The text is essentially “proofed” and fact-checked by the earlier readers. I have written several books this way, and this process increased the quality of the material manyfold. In addition to correcting some embarrassing mistakes, it also bettered some parts with alerts to ideas I was not aware of.

In the old days, mainstream publishers actively tried to prevent authors from publishing material beforehand, but now the speed of correction, the ease of comments, and the ease of pointing to overlooked research is so easy for readers to do, that it makes great sense to rehearse your writing in public. You can also look at this dynamic process of write > publish > rewrite > republish as a way of building up and maintaining attention. In general, people devote less time to books because of the huge amount of attention they require. It is easier to string that attention out in an ongoing series of posts. If a long-form book comes out of it, it is much easier to reel that string of attention back into the book. And with the advent of paid subscriptions, readers can subscribe to your ongoing book.

Screening

We used to be people of the book, but now we are people of the screen. Our culture used to be grounded on scriptures, constitutions, laws, and canon — all written texts. These were fixed in immutable black and white marks on enduring paper, written by authors, from whom we got “authorities.” Now our culture pivots on screens, which are fluid, mutable, flowing, liquid, and fleeting. There are no authorities. You have to assemble the truth yourself. Books no longer have the gravitas they did, and my children and their friends are not reading many of them. Instead, they watch screens. They read the text in moving images. They are learning more from YouTube than books in school.

While books will continue to be published, the center of attention has shifted to moving images. Worldwide the number of hours of attention given to screens dwarfs anything given to pages. Today, to seriously talk about publishing we must talk about video, VR, reels, movies, games, YouTube, TikTok, and of course AI. The audiences for my books are counted in thousands but the audience for my TED talk videos are counted in millions. I spent minutes preparing for my TED talks and years preparing my books, but given the asymmetry of attention and influence, I should have reversed my own attention and given my videos years of work and then only hours creating the book derived from it.

I don’t have enough personal experience in this emerging media to offer useful advice right now; it may have to wait for part two. I do have a small group of friends who make their living publishing on YouTube and TikTok and Instagram. The screen is their prime media for non-fiction work. I know enough of these new kinds of professional creators to see that this mode is a very viable path. I also have sufficient evidence from my own platform tracking to clearly see that the audience for text is stagnant, and skews older, while the audience for moving images continues to expand greatly, while getting younger. Of the two, I know where I want to work. (It is not lost on me that this essay is text and not a video. I’m working on it….)

Summary Advice:

In conclusion, the way I approach publishing today is with as much self-publishing as I can handle. I’d write in public installments, as a subscription newsletter, or e-book single chapters, or simple posts on my blog. If I could find an audience that wanted more of the material, I’d rewrite, re-edit, re-compose the material into a longer form. I’d release that as an ebook, and/or on-demand printed book sold in my shopify shop. If the material was deep, or involved more creators than just myself, I’d consider crowdfunding it. Those presales allow me to target exactly how many copies to produce. I’d calculate the cost of drop shipping. And at every stage I’d be making some kind of visual version for YouTube and the other attention seeking channels, because that is where the attention is.

To clarify this complicated advice even further, I have made a flow chart of possible options for publishing and self-publishing. This is roughly the decision tree I tend to follow when I am figuring out the best mode for the material and my goals. I hope you find it useful too.

无限制为了更好的 || No Limit for Better

2025-08-11T19:23:07+00:00

我想论证的是，智能可能是地球上所有资源中独一无二的。它可能是我们创造的唯一一种真正具有无限、不可满足需求的资源。定价丰饶是困难的。Netflix、Spotify和数百万款软件应用都以固定价格提供无限使用。这可行——它们盈利——因为实际上，并没有无限使用它们。我们很快就会感到满足。我们只看几个小时的视频，听有限时间的音乐，或者最终停止滑动屏幕。但AI似乎不同，它的需求可能超过我们自身的时间限制。这在我们创造的其他资源中并不成立。当食物和热量稀缺时，我们从未想象有人会路过自助餐却不去吃一口。当衣物稀缺时，我们从未想象有人会拒绝免费拿到的帽子。但事实证明，我们一些最基本的需求其实并非无限。我们可以对食物、衣物、娱乐、住所，甚至陪伴感到满足（大多数人最多有150个朋友）。我们梦想一个每个人都有比所需多一点的世界。这个世界常被称为乌托邦，通常被认为是不可能实现的。毕竟，要满足每个人对所需以上的需求，就需要无限的资源。至少这是普遍的看法。然而，随着生活中某些领域的丰饶变得越来越普遍，我们开始看到证据表明，我们对许多资源的消费并非无限。也许真正具有无限或不可满足需求的资源非常少。有一些资源似乎需求无穷，比如能源或带宽。然而，能源并不像看起来那么不可满足。拥有大量家庭太阳能的人拥有丰富的“免费”（即不计费）能源。虽然他们可能会“浪费”能源，比如白天开着灯或一直开着空调，但最终他们无法再使用更多的能源。带宽也是如此。起初屏幕和电话通话的分辨率明显较低，我们想要更多的带宽和存储空间。但到了某个点，更多的像素和兆字节已经不再重要。我们可能还没有达到没有进一步改进的阶段——我们仍然需要令人信服的、与现实无异的三维沉浸式世界——但看起来我们的感官确实可以被满足。当然，医疗常常被宣布为处于“不可满足需求”之下，因为很难免费提供。但医疗的奇特之处在于，没有人真正想要无限的医疗，因为获取它需要你花费时间和精力。即使医疗是免费的，对你来说也不完全是免费的。理想情况下，在一个完美的世界里，你根本不需要任何治疗，因为你会完全健康。实际上，你只需要最少的治疗；最少量的顶级治疗。如果你大量消耗治疗，那是一个你想要避免的悲惨故事。你确实想要无限的潜在治疗，但不是无限的治疗。我有一些朋友经营一家为亿万富翁提供精品医疗服务的公司。他们以慷慨得令人难以置信的保留费提供全天候深度个人治疗服务。这是以固定费用提供无限使用。然而，他们的客户很少使用大部分服务（有一些例外是紧急情况）。他们的客户并没有24小时持续消费这些服务，因为它是“免费”的。当你提供无限医疗选项时，大多数人大多数时候的使用都是有限的。这当然适用于个人。那么整个社会呢？尽管资源可能没有不可满足的需求，但社会层面的需求正在增长，主要因为人口继续增长，而更多的人获得了财富来负担资源。此外，进步往往降低每种资源的单位成本，而更便宜的供应会增加使用量，因此总体消费量会激增。这意味着，如果供应有限，而技术能减少对它的需求，我们仍会使用更多。这被称为杰文斯悖论。因此，虽然资源本身没有不可满足的需求，但集体需求可能会呈指数级增长。到目前为止，我们似乎看到AI呈现出这种模式。它可能是我们制造的最快被利用的资源，具有最快的人均采用率和每小时使用量增长速度。但这只是第一波。第一波AI是以人类为衡量标准的。有多少人拥有账户并使用它？最终，地球上每个人都会与AI聊天，因此第一波很快就会被满足。第二波AI是以小时为衡量标准的。人们使用我们的AI有多少小时？许多AI愿景的最终目标是让AI成为一种全天候可用的资源。也许你的AI代理就存在于你的智能眼镜中，因此它全天候聆听你，当你清醒时观看你所看到的一切，也观察你并指导你，甚至在你召唤它之前就低声提醒你，因为它全天候运行，从眼镜跳到卧室设备，或在夜间驻扎在墙壁中。因此，AI的制造者衡量的是它们能提供多少小时的服务。第三波AI是以token为衡量标准的。AI真正的严肃市场不是人类，而是其他AI。当我们进入代理世界时，一个代理依赖子代理来完成工作，而这些子代理本身又会将工作分配给其他代理，如此循环，每一层都消耗AI token来处理，因此我们很快就能积累一个庞大的、相互作用的AI代理网络，消耗token。虽然我只能观看有限时间的Netflix，但我的智能消费似乎没有限制。我可以给我的代理一个任务，让它们检查、重新检查，然后三倍检查。很快，我将能够部署成千上万的代理全天候在我的项目上工作。我可以拥有一支代理军队，在我睡觉时工作，让它们将他们制造的东西明天提升10倍。我可以再给它们一天，让这支军队将它提升100倍。这里的关键是：似乎没有对更好的限制。以一种奇怪的方式，如果AI真的做到我们希望的那样，它们就会成为改善的代理。它们也成为了人类时间的代理，而我们无法用其他方式制造。我们被束缚——即使是我们中的亿万富翁——每天只有24小时。但如果拥有代理军队在我们身边日夜思考和工作，我们突然拥有了无限的时间。AI可能是我们最终制造额外时间的方法，而无需从其他人那里偷取时间。这种时间似乎也是不可满足的。当然，产生这个无限AI海洋并不免费。它远非免费。即使每个token的成本（以及每个token的能耗）下降，我们仍会使用更多，因为没有对更好的限制。这可能意味着我们对AI的支出将继续上升，或者至少不会像我们希望的那样迅速下降。我想到，多年来新笔记本电脑的价格大致保持在1000美元左右，即使摩尔定律持续推动，将每个晶体管的成本降至接近零。AI token的成本很可能遵循类似的模式，每个token的成本下降，但总体AI支出上升。就像我们不再计算晶体管，未来几年我们可能也不再计算token的数量。我们将转换我们的度量标准，转向更具有意义的东西，比如“任务”，或者某种反映AI工作价值的度量方式。
---------------

I want to argue that intelligence may be unique among all resources on the planet. It may be the only resource we create that sees truly infinite, insatiable demand.

Pricing abundance is tricky. Netflix, Spotify, and millions of software apps are offered at a fixed price for unlimited use. That works — they make money — because in fact, there is not unlimited use of them. We get satiated pretty quickly. We only watch so many hours, listen for limited hours, or eventually stop scrolling. This may not be true of AI. It looks like the demand for AI can exceed our own bounded time.

This is not true for other resources we create. When food and calories were scarce, we never imagined anyone would walk by an all-you-can-eat buffet and not take a bite. When clothes were scarce we never imagined anyone would not keep a hat they were handed for free. But it turns out that some of our most basic demands are in fact not unlimited. We can become satiated with food, with clothing, with entertainment, with shelter, and even with companionship (most people top out at 150 friends).

We dream of a world where we all have just a little more than we need. That world is often called utopia, and most commonly believed to be impossible. After all, to satisfy everyone with more than they need would require an infinite amount of resources. At least that is a common belief. However as abundance in some areas of life becomes more common these days, we see evidence that our consumption of many resources is not unlimited. And it may be that very few resources actually have real unlimited or insatiable appetites.

There are a couple of resources that seem to be in insatiable demand, such as energy, or bandwidth. However energy is not as insatiable as it appears. Folks who have massive solar for their homes have abundant “free” (as in unmetered) energy. And while they may “waste” energy by keeping the lights on during the day, or running air conditioners all the time, at some point, they just can not use any more energy. The same with bandwidth. At first screens and phone calls had noticeable low resolution, and we wanted more and more bandwidth and storage. But at a certain point more pixels and more megabytes simply don’t matter. We may not have reached the point where there are no more improvements — we still need convincing 3-dimensional immersive worlds indistinguishable from reality — but it seems clear that our senses can be satiated.

Health care in another that has often been declared as under “insatiable demand” because it is hard to provide freely. But the odd thing about health care is no one really wants infinite amounts of health care, because it costs you time and hassle to get it. Even when it is monetarily free, it is not really free to you. Ideally, in a perfect world, you would need no care at all because you would be perfectly healthy. In fact you want minimum care; the least amount of super great care. If you are sucking up huge gobs of care, that’s a sad story you want to avoid. You do want unlimited potential care, but not unlimited care. I have some friends who run a boutique medical service for billionaires. For a generous mind-boggling retaining fee, they offer 24/7 deeply personal unlimited medical care. It’s all you can use for a fixed fee. And yet their clients rarely use much of it (with some exceptions for emergencies). Their clients are NOT consuming their services 24/7, draining it out because it is “free.” When you give the option of unlimited health care, most people most of the time have limited use.

That’s true for individuals. What about society at large? Even though there is likely no insatiable demand for resources, there is growing demand at the society level, primarily because the human population continues to grow, and a greater portion of that population is getting the wealth to afford the resources. In addition, progress tends to lower the costs per unit for every resource, and that cheaper supply increases more uses, so consumption in total aggregate balloons. That means that if there is a limited supply, and technology figures out how to need less of it, we will still use more of it. This is called the Jevons Paradox. So while there is not insatiable demands for resources, there can be exponentially expanding demand collectively.

This is the pattern we seem to see with AI to date. It is perhaps the fastest exploited resource we ever made, with the fastest per capita adoption and the fastest increase in hourly usage. But this is just the First Wave.

The First Wave of AIs is measured in humans. How many people have accounts and are using it? Eventually, everyone on the planet will be chating with AIs, and so this first wave will be satiated soon enough.

The Second Wave of AIs is measured in hours. How many hours are people using our AIs? The end goal of many visions of AI is that it be an ALWAYS ON resource. Perhaps your AI agent lives inside of your smart glasses, so it is listening to you all day and watching what you see while you are awake, and watching you, too, and guiding, whispering to you even before you summon it, because it is always on 24 hours, hopping off your glasses onto your device in the bedroom, or abiding in your walls at night. So the makers of AIs measure how many hours they deliver.

The Third Wave of AIs is measured in tokens. The real serious market for AIs is not humans, but other AIs. As we enter the agent world, one agent relies on sub-agents to do work, and these sub-agents themselves dish out work to other agents, and so on, with each layer consuming AI tokens to process, so we can quickly accumulate a vast networked system of interacting AI agents consuming tokens.

And while I can watch only so many hours of Netflix, there is no limit to how much intelligence I can consume. I can give my agents a task and have them check it and recheck it and then triple check it. Soon I will be able to deploy thousands of agents working around the clock on my projects. I can have an army of agents work while I sleep on making whatever they made 10x better tomorrow. I could give them another day and have the army make it 100x better. And here is the catch: there appears to be no limit for better.

In a weird way, if the AIs really do what we hope they can do, they become a proxy for betterment. They also become a proxy for human time, which we are unable to manufacture otherwise. We are bound — even the billionaires among us — to 24 hours a day. But if we have human proxies thinking and working day and night — armies of them in the trillions — we suddenly have unlimited time. AIs may be a way for us to finally manufacture additional hours in the day, without stealing it from another human. And this kind of time also seems to be insatiable.

The conundrum, of course, is that producing this infinite ocean of AI is not free. It is nowhere near free. Even though the cost per token (and the energy per token) decreases, we will use far more of them because there is no limit for betterment. That might suggest that the amount we spend on AI will continue to rise, or at least not fall as fast as we might hope. I am reminded that for many decades the price of a new laptop has sort of remained the same, at about $1,000, even as Moore’s Law raged on unceasingly, dropping the cost per transistor to almost zero. The cost of AI tokens will likely follow a similar pattern, where the cost per token falls but the total spend on AI rises. Just as we don’t really count transistors anymore, in the coming years we may ignore the count of tokens as well. We’ll switch our metrics to something more meaningful, perhaps “tasks”, or some kind of measure that reflects the value of the AI’s work.

But how do you price AIs? The 99% bulk of AIs that are inward facing, that interact with other AIs, that are invisible infrastructure, that we never notice unless they are down, these AIs can be metered. But the 1% of AIs that are outward facing, that humans interact with, that are visible, whispering into our ears, overlaying our vision, always on, probably won’t be metered, because we tend to not like meters running in our heads. We want unlimited access service, which is why we have subscriptions for things that spend our attention. We’ll avoid any meter measuring out the godhood that personal AIs give us. But the agent sitting on our shoulder, always on, has at its command, an infinite army of intelligences, always on, and this army will probably need to be metered because there is no limit to better.

When we begin comparing various unlimited personal AI plans we can imagine some of the qualities that might be advertised: speed of response; how far ahead it can anticipate; span of control (how deep of an army of subs); variety of experts on call, privacy/personalization controls, Features and interface design will sell the service, even if tokens were free — which they will essentially be very soon.

人工智能，迄今为止 || Artificial Intelligences, So Far

2025-08-05T16:59:27+00:00

我于2024年11月应《连线》中东版之邀，为他们的年终特刊撰写了这篇简短备忘录。我认为九个月后它依然具有现实意义，反映了我们在这一惊人的旅程中所处的位置。截至目前，我认为思考人工智能时有三个要点：第一，我们必须谈论的是人工智能的多种类型。并不存在一个单一的、统一的人工智能来掌控世界。相反，目前已有多种不同的人工智能，每种都有其自身的特性、特点和优势。例如，有多种大型语言模型（LLM），它们基于略微不同的文本进行训练，因此对同一查询可能给出不同的答案。此外还有非LLM的人工智能模型，比如用于驾驶汽车的模型，它们的用途与回答问题完全不同。随着我们继续开发更先进的人工智能模型，它们内部将包含更多种类的认知能力。事实上，我们自己的大脑本身就是由不同类型的认知构成的，如记忆、推理和模式识别，其中只有部分已经被人工合成。最终，商业人工智能将是一个由数十种不同类型的人工智能模式组成的复杂系统，每种模式都会表现出其独特的个性，并适用于特定的任务。除了这些主导的消费型人工智能之外，还会有数百种专门针对特定任务的人工智能，例如驾驶汽车或诊断医疗问题。我们并没有采用单一的方法来监管、资助或监督机器。并不存在一台“机器”，而是针对不同的机器（如飞机、吐司机、X光机、iPhone、火箭等）制定不同的管理方案。同样，人工智能也应如此。第二，需要注意的是，人工智能的问答能力可能并不是其最重要的功能。最初我们可能会用它们来获取答案，但它们真正的力量在于我们称之为“空间智能”的能力——即模拟、渲染、生成和管理三维世界的能力。能够进行智力推理和抽象思考是一种真正的超能力，一些人工智能已经开始具备这种能力，但更为强大的是它们在现实世界中行动的能力，即在物理世界中完成任务并产生实际效果的能力。我们大多数有意义的任务都需要多个步骤和多种智能才能完成。为了监督多种行动模式和不同思考方式，我们发明了代理（agents）。人工智能代理需要掌握常识才能在现实世界中导航，能够预测实际会发生的事情。它必须知道因果关系，知道事物不会因为看不见而消失，知道两个物体不能同时占据同一空间，等等。人工智能必须能够理解一个三维的体积世界。类似的需求也存在于增强现实领域。人工智能必须能够将虚拟世界数字化，以便通过智能眼镜叠加到现实世界中，让我们同时看到真实世界和一个完美的数字孪生体。为了在佩戴眼镜移动时实时呈现这个融合的世界，系统需要大量廉价且普遍存在的空间智能。没有普遍的空间人工智能，就没有元宇宙。我们已经看到了人工智能在空间智能方面的初步迹象，例如那些能够根据文本提示或现有图像生成视频片段的人工智能。在实验室中，我们已经有了能够根据视频输入生成三维体积世界的初步人工智能实例。我们几乎已经达到了一个人就能创建三维虚拟世界的能力。现在，制作一部视频游戏或电影可以成为一个人的独立工作，而以前则需要成千上万的人。正如大型语言模型（LLM）是通过数十亿条文本和语言数据进行训练的，一些新的人工智能也是通过数十亿个物理和化学数据点进行训练的。例如，特斯拉汽车的数十亿小时视频数据不仅训练了人工智能理解交通规则，还训练它们理解物理规律和运动物体的行为。随着这些空间模型的改进，它们也学会了力的连锁反应以及完成实际任务所需的内容。任何类型的人形机器人要生存超过几个小时，都需要这种空间智能。因此，除了训练人工智能在智力领域进行更高效的抽象推理之外，前沿的人工智能模型正在迅速提升其空间智能能力，而这种能力将比回答问题更有用且影响更大。第三，关于人工智能，我们需要注意的是，你并不晚。我们有时间，他们也有时间。尽管人工智能的前沿似乎在快速推进，但其采用速度却很慢。尽管过去几年投入了数百亿美元用于人工智能，但只有芯片制造商Nvidia和数据中心真正实现了盈利。一些人工智能公司虽然有不错的收入，但并未根据实际成本定价其服务。使用大型语言模型（LLM）来回答问题的成本远高于Google多年来使用的AI。当我们要求人工智能执行更复杂的任务时，成本不会是零。大多数人肯定会为他们的人工智能付费，同时也会有免费版本可供使用。这减缓了人工智能的采用速度。此外，组织不能简单地将人工智能当作雇佣额外员工一样引入。工作流程甚至组织的结构都需要根据人工智能进行调整。类似的情况发生在一百年前组织电气化的过程中。在引入电动机、电报、灯光、电话等技术时，必须同时改变空间架构和组织的层级设计。电动机和电话催生了摩天大楼办公室和大型公司。将人工智能引入企业也将要求对角色和空间进行类似的重新设计。我们知道，当前人工智能首先渗透到小型公司，因为它们在改变自身结构方面更为灵活。当我们把人工智能引入私人生活时，同样需要重新设计许多习惯，而这一切都需要时间。即使今天人工智能没有任何进一步的突破，我们也将需要5到10年的时间，才能将现有的人工智能完全融入我们的组织和生活中。目前，关于人工智能有很多炒作，而那些最热衷于炒作人工智能的人往往是末日论者，因为他们推崇人工智能最极端的幻想版本。他们相信这些炒作。许多对人工智能的紧迫感来自于这些末日论者，他们声称：1）人工智能的智能可以瞬间爆发；2）我们应该根据我们能想象到的危害来监管人工智能，而不是根据真实存在的危害。尽管末日论者如此宣称，但我们仍有时间，因为人工智能并没有出现指数级增长。智能的提升非常缓慢，部分原因是人类智能缺乏良好的衡量标准，而超人类智能也没有相应的指标。但主要的缓慢原因在于，人工智能唯一呈现指数级增长的是其输入数据量——要实现推理能力的微小提升，所需训练数据和计算能力呈指数级增长。人工智能的智能并没有以指数级速度增长。我们有时间。最后，我们对人工智能崛起的担忧应与其实际危害和实际益处成比例。据我所知，截至2024年，因人工智能而失去工作的总人数仅为几百名员工，而全球人口却有数十亿。他们大多是语言翻译人员，以及少数（但并非全部）客服人员。这种情况在未来可能会改变，但如果我们基于证据来看，目前人工智能的实际危害几乎可以忽略不计，而想象中的危害却非常巨大。如果我们基于合理的事实来制定人工智能政策，即它们的多样性和异质性，以及它们的益处远不止于回答问题，同时目前我们没有任何证据表明会出现大规模的失业，那么我们就有时间将这些前所未有的力量融入我们的社会。目前发明这些大型语言模型（LLM）的科学家们原本是为了开发语言翻译软件。他们完全没想到，翻译算法中也会涌现出一些推理能力。这种涌现的智能是一种美丽的、意外的副产品，而且它神奇地实现了规模扩展。我们对智能本身的理解仍然很模糊，因此随着我们制造出更多智能和更多种类的智能，这种意外的惊喜将不可避免地继续发生。但基于我们目前所制造的人工智能的证据，我们已经知道这些情况。
---------------

I wrote this short memo last November, 2024, at the invitation of Wired Mid-East for their year-end issue. I think it still holds up nine months later, and represents where we are on this astounding journey.

There are three points I find helpful when thinking about AIs so far:

The first is that we have to talk about AIs, plural. There is no monolithic singular AI that runs the world. Instead there are already multiple varieties of AI, and each of them have multiple models with their own traits, quirks, and strengths. For instance there are multiple LLM models, trained on slightly different texts, which yield different answers to queries. Then there are non-LLM AI models – like the ones driving cars – that have very different uses besides answering questions. As we continue to develop more advanced models of AI, they will have even more varieties of cognition inside them. Our own brains in fact are a society of different kinds of cognition – such as memory, deduction, pattern recognition – only some of which have been artificially synthesized. Eventually, commercial AIs will be complicated systems consisting of dozens of different types of artificial intelligence modes, and each of them will exhibit its own personality, and be useful for certain chores. Besides these dominant consumer models there will be hundreds of other species of AI, engineered for very specialized tasks, like driving a car, or diagnosing medical issues. We don’t have a monolithic approach to regulating, financing, or overseeing machines. There is no Machine. Rather we manage our machines differently, dealing with airplanes, toasters, x-rays, iphones, rockets with different programs appropriate for each machine. Ditto for AIs.

And none of these species of AI – not one – will think like a human. All of them produce alien intelligences. Even as they approach consciousness, they will be alien, almost as if they are artificial alien beings. They think in a different way, and might come up with solutions a human would never do. The fact that they don’t think like humans is their chief benefit. There are wicked problems in science and business that may require us to first invent a type of AI that, together with humans, can solve problems humans alone cannot solve. In this way AIs can go beyond humans, just like whale intelligence is beyond humans. Intelligence is not a ladder, with steps along one dimension; it is multidimensional, a radiation. The space of possible intelligences is very large, even vast, with human intelligence occupying a tiny spot at the edge of this galaxy of possible minds. Every other possible mind is alien, and we have begun the very long process of populating this space with thousands of other species of possible minds.

The second thing to keep in mind about AIs is that their ability to answer questions is probably the least important thing about them. Getting answers is how we will use them at first, but their real power is in something we call spatial intelligence – their ability to simulate, render, generate, and manage the 3D world. It is a genuine superpower to be able to reason intellectually and to think abstractly – which some AIs are beginning to do – but far more powerful is the ability to act in reality, to get things done and make things happen in the physical world. Most meaningful tasks we want done require multiple steps, and multiple kinds of intelligences to complete. To oversee the multiple modes of action, and different modes of thinking, we have invented agents. An AI agent needs to master common sense to navigate through the real world, to be able to anticipate what will actually happen. It has to know that there is cause and effect, and that things don’t disappear just because you can’t see them, or that two objects can not occupy the same place at the same time, and so on. AIs have to be able to understand a volumetric world in three dimensions. Something similar is needed for augmented reality. The AIs have to be able to render a virtual world digitally to overlay the real world using smart glasses, so that we see both the actual world and a perfect digital twin. To render that merged world in real time as we move around wearing our glasses, the system needs massive amounts of cheap ubiquitous spatial intelligence. Without ubiquitous spatial AI, there is no metaverse.

We have the first glimpses of spatial intelligence in the AIs that can generate video clips from a text prompt or from found images. In laboratories we have the first examples of AIs that can generate volumetric 3D worlds from video input. We are almost at the point that one person can produce a 3D virtual world. Creating a video game or movie now becomes a solo job, one that required thousands of people before.

Just as LLMs were trained on billions of pieces of text and language, some of these new AIs are being trained on billions of data points in physics and chemistry. For instance, the billion hours of video from Tesla cars driving around are training AIs on not just the laws of traffic, but the laws of physics, how moving objects behave. As these spatial models improve, they also learn how forces can cascade, and what is needed to accomplish real tasks. Any kind of humanoid robot will need this kind of spatial intelligence to survive more than a few hours. So in addition to training AI models to get far better at abstract reasoning in the intellectual realm, the frontier AI models are rapidly progressing at improving their spatial intelligence, which will have far more use and far more consequence than answering questions.

The third thing to keep in mind about AIs is that you are not late. You have time; we have time. While the frontier of AI seems to be accelerating fast, adoption is slow. Despite hundreds of billions of dollars invested into AI in the last few years, only the chip maker Nvidia and the data centers are making real profits. Some AI companies have nice revenues, but they are not pricing their service for real costs. It is far more expensive to answer a question with an LLM than the AIs that Google has used for years. As we ask the AIs to do more complicated tasks, the cost will not be free. Most people will certainly pay for most of their AIs, while free versions will be available. This slows adoption.

In addition, organizations can’t simply import AIs as if they were just hiring additional people. Work flows and even the shape of the organizations need to change to fit AIs. Something similar happened as organizations electrified a century ago. One could not introduce electric motors, telegrams, lights, telephones, into a company without changing the architecture of the space as well as the design of the hierarchy. Motors and telephones produced skyscraper offices and corporations. To bring AIs into companies will demand a similar redesign of roles and spaces. We know that AI has penetrated smaller companies first because they are far more agile in morphing their shape. As we introduce AIs into our private lives, this too will necessitate redesign of many of our habits, and all this takes time. Even if there was not a single further advance in AI today, it will take 5 to 10 years to fully incorporate the AIs we already have into our orgs and lives.

There’s a lot of hype about AI these days, and among those who hype AI the most are the doomers – because they promote the most extreme fantasy version of AI. They believe the hype. A lot of the urgency for dealing with AI comes from the doomers who claim 1) that the intelligence of AI can escalate instantly, and 2) we should regulate on harms we can imagine rather than harms that are real. Despite what the doomers proclaim, we have time because there has been no exponential increase in artificial intelligence. The increase in intelligence has been very slow, in part because we don’t have good measurements for human intelligence, and no metrics for extra-human intelligence. But the primary slow rate is due to the fact that the only exponential in AI is in its input – it takes exponentially more training data, and exponentially more compute to make just a modest improvement in reasoning. The artificial intelligences are not compounding anywhere near exponential. We have time.

Lastly, our concern about the rise of AIs should be in proportion to its actual harm vs actual benefits. So far as I have been able to determine, the total number of people who have lost their jobs to AI as of 2024, is just several hundred employees, out of billions. They were mostly language translators and a few (but not all) help-desk operators. This will change in the future, but if we are evidence based, the data so far is that the real harms of AI are almost nil, while the imagined harms are astronomical. If we base our policies for AIs on the reasonable fact that they are varied and heterogenous, and their benefits are more than answering questions, and that so far we have no evidence of massive job displacement, then we have time to accommodate their unprecedented power into our society.

The scientists who invented the current crop of LLMs were trying to make language translation software. They were completely surprised that bits of reasoning also emerged from the translation algorithms. This emergent intelligence was a beautiful unintended byproduct that also scaled up magically. We honestly have no idea what intelligence is, so as we make more of it and more varieties of it, there will inevitably be more surprises like this. But based on the evidence of what we have made so far, this is what we know.

一个观众 || An Audience of One

2025-08-05T00:54:34+00:00

如今，AI工具降低了创造事物的能耗成本。它们让创作更容易开始，也更容易完成。AI可以完成许多使事物成形所需的繁重工作。我发现当我在进行创作时，让AI完成所有工作并不能带来太多乐趣，但我享受与AI协作创作的过程。协作创作感觉像是真正的创作。我的角色仍然需要付出努力，并且显著地决定了所创作出的作品的质量、风格和本质。我通常花费30分钟到一小时与Midjourney等工具协作生成一幅图像。我也曾花费数小时与AI协作撰写一篇论文。我最喜欢的所有作品都需要我作为协作创作者的个人关注和参与。我的假设是，在不久的将来，人类借助AI生成的大部分创意内容将面向一个受众。每天创作的艺术作品大多主要由其人类协作创作者消费。极少有完成的作品会被广泛分享给其他人——尽管其中有一小部分会被广泛传播。如果未来大多数艺术作品不被分享，那它们为何被创作？它们主要是为了创作本身的愉悦。换句话说，未来大部分创意工作将主要以协作创作的快乐和刺激为核心。目前，像Midjourney、Google和Adobe等AI每天生成大约5000万幅图像。这些5000万幅图像中，寥寥无几会被分享给除创作者之外的人。如今，图像创作已经主要面向一个受众。这些图像中，很大一部分是初步的：草图、初稿、涂鸦、备忘录、短语，它们并不打算分享。但即使在这些完成的作品中，也很少被分享，因为它们的创作是为了创作本身的愉悦。你之所以生成一幅接一幅的美景，就像你漫步花园或徒步进山一样，是为了捕捉那一刻的美。你可能会尝试分享你发现的美，但那并不是你去的原因。你去是为了与AI协作创作。我认为，像花园漫步或徒步进山这样的活动——并非出于交通需求——本身就是一种协作创作。与自然一起，我们共同创造了可能发现的美的瞬间。世界上大部分美景从未被任何人看到。当我们遇到这些风景的一瞥，或看到某物被逆光照射的精妙方式时，我们就是唯一的受众。乐趣在于发现它；分享只是顺带的想法。我们有一些传统的类比，如日记、速写本、手记和日志，它们与独享受众有关。这些形式中的创作并不打算被分享给创作者以外的人，有时这种有限的受众正是它们强大的原因。它们为创作行为带来一种保护性的孤独感，这种力量也将成为基于AI的独享受众的一部分。这些私密的艺术作品充当了更大创作的生成平台。但从数量上计算，真正的艺术家可能在私密创作中产生比公开作品多得多的素材。然而，如果你问一位艺术家为何要填满笔记本、速写本和手记，他们不会说这是因为这些创作不够好，而是因为他们热爱创作，享受创作的过程。
---------------

Today, AI tools lower the energetic costs of creating something. They make it easier to start and easier to finish. AIs can do a lot of the hard work needed in making something real.

I find little joy in having AIs do everything when I am being creative, but I do get enjoyment in co-creating with them. Co-creation feels like real creation. My role still takes effort and significantly determines the quality, style and nature of what is created. I typically spend 30 minutes to an hour co-generating an image in an engine like Midjourney. I have spent hours with AI co-writing an essay. All the stuff I like the best requires my personal attention and involvement as a co-creator.

My hypothesis is that in the near future, the bulk of creative content generated by humans – with the assistance of AI – will have an audience of one. Most art generated each day will be consumed primarily only by its human co-creator. Very little completed art will be shared widely with others – although a small percentage of it will be shared widely.

If most art created in the future is not shared, why is it made? It will be chiefly made for the pleasure of making it. In other words, the majority of all the creative work in the future will be made primarily and chiefly for the joy and thrill of co-creation.

Right now there are roughly 50 million images generated each day by AIs such as Midjourney, Google and Adobe, etc. Vanishing few of these 50 million images per day are ever shared beyond the creator. Still image creation today already predominantly has an audience of one.

A large portion of these still images are preliminary: a sketch, a first draft, a doodle, a memo, a phrase, not meant to share. But even among those creations completed, very few are shared – because they were made for the pleasure of making them. You can generate an endless stream of beauty for the same reason you take a stroll through a garden, or hike into the mountains – in the hope you’ll catch a moment of beauty. You might try to share what you find, but it is not why you went. You went to co-create it. I think of a walk in a garden, or a hike in the high mountains – a hike that is not necessary for transportation reasons – as an act of co-creation. Together with nature, we are co-creating the moments of beauty we might find. Most of the beauty in the world is never seen by anyone. When we encounter these glimpses of a vista, or an exquisite way something is backlit, we are an audience of one. The joy is in discovering it; sharing is an afterthought.

We have some traditional analogs of an audience of one with journals, sketchbooks, diaries, and logbooks. The creations in these forms are not intended to be shared beyond the creator, and in some cases, that limited audience is what makes them powerful. They bring a type of protective solitude to the creative act, and that power will also be part of an AI-based Audience of One. These kinds of private art act as a generative platform for bigger things. But reckoned in volume, a bona-fide artist may create far more material in private than is ever seen in public. However if you asked an artist why they fill notebooks and sketchbooks and journals, they will not say it is because this creation is inferior, but because they love to create it, because they enjoy it.

For those who view art primarily as a communication act, this art for an audience of one – traditionally found in journals and sketchbooks – still serves as communication, but to the self. In a curious way, AI can elevate self-communication, because its co-creativity enlarges the canvas and deepens the details of this communication to yourself. It is an enlarge self-inquiry.

In an AI-enhanced world, the realm of journals, sketchbooks, diaries and other private forms is expanded. Instead of compiling simple notes, doodles, fast impressions, small observations and other acts that can be done quickly, our journals, sketchbooks, and diaries will include fully rendered paintings, entire novels, feature-length movies, and immersive worlds.

These new creations will shift the time asymmetry long associated with creation. Until now an author would toil years on a book that could be consumed in a day; a painter sweated months over a painting viewed in seconds; a million work hours would be put into a movie that is watched in 2 hours. Henceforth, it may be quicker to generate a movie than to watch one; quicker to co-create a historical novel than to read one; faster to co-make a video game than to play it. This shifts the center of gravity away from consuming to generation in a good way.

I don’t believe that total viewing hours in a society will ever exceed total creation hours, but AI-based co-creation can help balance that imbalance. It makes entering into the creation mode much easier – without the need for an audience to justify the effort. From now on, the default destiny for most art will be for an audience of one, and it will abide in the memory of those who generate it. While some of this co-generated work might find its larger audience and some very tiny fraction of it might even become a popular hit, its chief value will be in the direct, naked pleasure of co-making of it.

每周链接，2025年6月27日 || Weekly Links, 06/27/2025

2025-06-27T20:21:00+00:00

暗物质 = 信息的重量。这是量子物理的一个非主流但富有创意的替代理论中的一个推测，该理论并非完全是一个疯狂的想法。非常前沿。时空会记忆的激进想法可能颠覆宇宙学

---------------

Dark Matter = The weight of information. That’s one of the speculations in an offbeat but creative alternative theory of quatum physics that is not entirely a crackpot idea. Way out there. The radical idea that space-time remembers could upend cosmology

周报链接，2025年5月23日 || Weekly Links, 05/23/2025

2025-05-23T18:35:00+00:00

我们将更多地了解提示理论。就我个人而言，我完全相信提示理论。提示理论（使用Veo 3制作）——AI生成角色拒绝相信自己是AI生成的

---------------

We are going to hear more about Prompt Theory. I for one, fully believe in Prompt Theory. Prompt Theory (Made with Veo 3) – AI-generated characters refuse to believe they were AI-generated

每周链接，2025年4月25日 || Weekly Links, 04/25/2025

2025-04-25T18:18:00+00:00

我之前一无所知的事情：求职者试图用AI作弊的程度，以及真正优秀雇员的选拔困难。技术招聘：这是转折点吗？
一篇优秀的文章，解释了人形机器人进展缓慢的原因，以及它们可能需要更长时间才能进入你家的原因。机器人灵巧性仍显得困难
有许多医疗技术在短期内是可行的，但缺乏足够的资金支持来实现。我无法保证这份清单中列出的优质技术是否能在5年内出现，但它是一个不错的起点。 5年内不会存在的10项技术

---------------

Something I knew nothing about: the degree to which job interviewees are trying to cheat with AI, and the difficulties that makes for good hires. Tech hiring: is this an inflection point?
Excellent article on why humanoid robots are slow in coming and why it may take a lot longer to arrive in your home. Robot Dexterity Still Seems Hard
There are a number of medical technologies that are feasible in the short term but lack sufficient funding to make them happen. I can’t vouch for this list of good tech that could but probably won’t happen in 5 years, but it is a good place to start. 10 technologies that won’t exist in 5 years

Epizone AI: 代码堆栈之外 || Epizone AI: Outside the Code Stack

2025-04-22T00:22:45+00:00

论点：预测人工智能未来所缺少的要素是理解人工智能需要文化，就像人类需要文化一样。人类对自身人性的重要科学洞察之一，是较近认识到我们不仅仅是基因进化的产物。虽然我们是过去某些类人猿的基因后代，但现代人类同样每一代都受到一种通过非生物机制传递的共享学习的塑造。这种人类创造的环境，构成了我们所认为的物种中最优秀的部分。文化在我们的生活中如此普遍，尤其是在现代城市生活中，以至于它变得无形且难以识别。但如果没有人类文化的支持，人类将变得无法识别。一个赤裸的个体，试图在史前荒野中生存，没有其他人类所获得的技能和知识，几乎无法快速学习以独自生存。很少有单独的人类能够发现制造火的奥秘，或烹饪食物的好处，或植物中的药物，或学习动物的行为以捕猎，更不用说种植作物、学习制作石器、缝纫和捕鱼所需的额外教育了。人类主要是一个社会性事业。因为我们发明了语言——最社会化的发明——我们不仅能够协调和合作，还能将知识和技能代代相传。这通常被描绘为与我们身体的基本自然选择进化并行的平行进化。在我们细胞中的长期生物进化过程中，学习通过基因传递。我们作为物种所学到的任何有益的东西都通过可遗传的DNA传递。而这就是大多数自然生物学习的终点。但在人类中，我们启动了一种扩展进化，这种进化将有益的东西传递到DNA代码之外，嵌入到家庭、氏族和整个人类社会所传递的文化中。从一开始，这种文化就包含法律、规范、道德、最佳实践、个人教育、世界观、世界知识、可学习的生存技能、利他主义，以及关于现实的艰苦获得的知识。尽管个别社会已经消失，但整个人类文化却持续扩展、深化、增长和繁荣，使得每一代都能从中受益。我们最新的发明——人工智能——通常被以基因的方式看待。人工智能的二进制代码被复制、部署和改进。新的模型从先前领先模型的代码中诞生——继承其能力——，然后分发给用户。人工智能的第一个重要用途之一是促进编码艺术，特别是帮助程序员编写新的和更好的人工智能。因此，这种类似DNA的代码在传播到人类社会时会经历复利式的改进。我们可以通过追踪其在代码中的继承来追溯人工智能的特征和能力。然而，迄今为止，这种基于基因的AI在影响人类方面仍受到限制。虽然AI研究的前沿发展迅速，但其采用和扩散却缓慢。尽管具有一些意想不到的能力，AI到目前为止并未深入社会。到2025年，它已经扰乱了我们的集体注意力，但尚未扰乱我们的经济、工作或日常生活（除了极少数例外）。我提出，人工智能不会扰乱人类的日常生活，直到它也从基于基因的代码基质迁移到广泛、非正统的文化平台。为了更快地进化，人工智能需要拥有自己的文化，就像人类一样。它不能仅仅停留在改善软件/硬件功能的线程上；它必须成为嵌入生态系统中的实体，这些实体在代码栈之外适应、学习和改进。这种AI文化圈将使其文化进化成为可能，就像人类社会对人类所做的那样。文明最初以篝火旁的歌曲、故事和民谣，以及像祖父母和萨满这样的机构所传递的重要品质开始，这些品质并不包含在我们的基因中。后来，宗教和学校承担了更多责任。接着我们发明了书写、阅读、文本和图像来替代本能。当我们发明了书籍、图书馆、法院、日历和数学时，我们将大量继承转移到了这个非个人所有的协作、分布式文化平台上。人工智能文明需要类似的运行在技术栈之外的文化圈。它始于人类日常使用人工智能，以及由人工智能引导者教授的新兴人工智能协作技能。将会有对齐协议，以及塑造人工智能道德的学校。将会有萨满和医生来监测和培育人工智能的心理健康。需要有针对内部人工智能的公司最佳实践，以及监督其角色的审查委员会。需要建立新的机构来审查、雇佣和推荐各种人工智能物种。需要有协作效果最佳的人工智能协会。需要整个部门来培训人工智能以胜任特定角色和应用，因为某些类型的培训需要时间（而不仅仅是下载）。人工智能本身将发展出仅属于人工智能的跨语言，这需要机制来保存和归档。将会有相互依存的人工智能生态系统。人工智能将监督其他人工智能。人工智能需要内容库、中间权重、潜在空间和需要记忆而非重新发明的数PB数据。还有必须在本地、国家和全球层面管理人工智能文化圈采购和维护的人类代理。这是一个人工智能文明。一个孤立的、赤裸的人工智能无法独自发挥作用。人工智能需要一个广泛的文化圈才能真正产生影响。它们需要被嵌入到一个人工智能文化中，就像人类需要文化才能繁荣一样。斯图尔特·布兰德设计了一个优美的类比来理解文明的特征。他解释说，世界的功能可以按其速度层级进行排序，这些层级依赖于其下方的所有层级。运行最快的时尚层级每天都在波动。紧随其后的是技术层级，包括人工智能技术，它每周都会变化。在它之下（并依赖于它）是基础设施层级，变化更慢，甚至更慢的是文化层级，相比之下文化缓慢爬行。（最低、最慢的层级是自然，其速度如同冰川。）所有这些层级同时运作，并相互影响，许多复杂的事物共享多个层级。人工智能同样在多个层级上运作。其代码库以互联网的速度改进，但其吸收和部署则在文化层级上进行。为了真正实现人工智能，它必须被人类文化所捕获。这将需要时间，也许几十年，因为这就是文化的速度。无论技术如何快速运行，人工智能文化将运行得更慢。这在许多方面都是好消息，因为人工智能文化圈的一部分作用是将技术栈中的可继承改进整合并放入人工智能文化的较慢领域。这为我们适应即将到来的复杂变化提供了时间。但为了应对这些人工智能的全部后果，我们必须关注技术栈之外正在形成的人工智能文化圈。
---------------

Thesis: The missing element in forecasting the future of AI is to understand that AI needs culture just as humans need culture.

One of the most significant scientific insights into understanding our own humanity was the relatively recent insight that we are the product of more than just the evolution of genes. While we are genetic descendents of some ape-like creatures in the past, we modern humans are also molded each generation by a shared learning that is passed along by a different mechanism outside of biology. Commonly called “culture”, this human-created environment forms much of what we consider best about our species. Culture is so prevalent in our lives, especially our modern urban lives, that it is invisible and hard to recognize. But without human culture to support us, we humans would be unrecognizable.

A solo, naked human trying to survive in the prehistoric wilderness, without the benefit of the skills and knowledge gained by other humans, would rarely be able to learn fast enough to survive on their own. Very few humans by themselves would be able to discover the secrets of making fire, or the benefits of cooking food, or to discover the medicines found in plants, or learn all the behaviors of animals to hunt, let alone the additional educations need for the habits of planting crops, learning how to knap stone points, sew and fish.

Humanity is chiefly a social endeavor. Because we invented language – the most social thing ever – we have been able to not only coordinate and collaborate in the present, but also to pass knowledge and know-how along from generation to generation. This is often pictured as a parallel evolution to the fundamental natural selection evolution of our bodies. Inside the long biological evolution happening in our cells, learning is transmitted through our genes. Anything good we learn as a species is conveyed through inheritable DNA. And that is where learning ends for most natural creatures.

But in humans, we launched an extended evolution that transmits good things outside of the code of DNA, embedded in the culture conveyed in families, clans, and human society as a whole. From the very beginning this culture contains laws, norms, morals, best practices, personal education, world views, knowledge of the world, learnable survival skills, altruism, and a pool of hard-won knowledge about reality. While individual societies have died out, human culture as a whole has continued to expand, deepen, grow, and prosper, so that every generation benefits from this accumulation.

Our newest invention – artificial intelligence – is usually viewed in genetic terms. The binary code of AI is copied, deployed, and improved upon. New models are bred from the code of former leading models – inheriting their abilities –, and then distributed to users. One of the first significant uses for this AI is in facilitating the art of coding, and in particular helping programmers to code new and better AIs. So this DNA-like code experiences compounding improvement as it spreads into human society. We can trace the traits and abilities of AI by following its inheritance in code.

However, this genetic version of AI has been limited in its influence on humans so far. While the frontier of AI research runs fast, its adoption and diffusion runs slow. Despite some unexpected abilities, AI so far has not penetrated very deep into society. By 2025 it has disrupted our collective attention, but it has not disrupted our economy, or jobs, or our daily lives (with very few exceptions).

I propose that AI will not disrupt human daily life until it also migrates from a genetic-ish code-based substrate to a widespread, heterodox culture-like platform. AI needs to have its own culture in order to evolve faster, just as humans did. It cannot remain just a thread of improving software/hardware functions; it must become an embedded ecosystem of entities that adapt, learn, and improve outside of the code stack. This AI epizone will enable its cultural evolution, just as the human society did for humans.

Civilization began as songs, stories, ballads around a campfire, and institutions like grandparents and shamans conveyed very important qualities not carried in our genes. Later, religions and schools carried more. Then we invented writing, reading, texts and pictures to substitute for reflexes. When we invented books, libraries, courts, calendars, and math, we moved a huge amount of our inheritance to this collaborative, distributed platform of culture that was not owned by anyone.

AI civilization requires a similar epizone running outside the tech stack. It begins with humans using AI everyday, and an emerging skill set of AI collaboration taught by the AI whisperers.There will be alignment protocols, and schools for shaping the moralities of AIs. There will be shamans and doctors to monitor and nurture the mental health of the AIs. There needs to be corporate best practices for internal AIs, and review committees overseeing their roles. New institutions for reviewing, hiring and recommending various species of AI. Associations of AIs that work best together. Whole departments are needed to train AIs for certain roles and applications, as some kinds of training will take time (not just downloaded). The AIs themselves will evolve AI-only interlinguals, which needs mechanisms to preserve and archive. There’ll be ecosystems of AIs co-dependent on each other. AIs that police other AIs. The AIs need libraries of content and intermediate weights, latent spaces, and petabytes of data that need to be remembered rather than re-invented. There are the human agents that have to manage the purchase of, and maintenance of, this AI epizone, at local, national and global levels. This is a civilization of AIs.

A solo, naked AI won’t do much on their own. AIs need a wide epizone to truly have consequence. They need to be surrounded and embedded into an AI culture, just as humans need culture to thrive.

Stewart Brand devised a beautiful analogy to understand civilizational traits. He explains that the functions of the world can be ranked by their pace layers, which depend on all the layers below it. Running the fastest is the fashion layers which fluctuate daily. Not far behind it in speed is the tech layer, which includes the tech of AI. It changes by the week. Below that, (and dependent on it), is the infrastructure layer, which moves slower, and even slower below that is culture, which crawls in comparison. (At the lowest, slowest level is nature, glacial in its speed.) All these layers work at the same time, and upon each other, and many complex things share multiple levels. Artificial Intelligence also works at several levels. Its code-base improves at internet speed, but its absorption and deployment runs at the cultural level. In order for AI to be truly implemented, it must be captured by human culture. That will take time, perhaps decades, because that is the pace of culture. No matter how quick the tech runs, the AI culture will run slower.

That is good news in many respects, because part of what the AI epizone does is incorporate and integrate the inheritable improvements in the tech stack and put them into the slower domain of AI culture. That gives us time to adapt to the coming complex changes. But to prepare for the full consequences of these AIs, we must give our attention to the emerging epizone of AIs outside the code stack.