Why I am against GenAI and everything it stands for

为什么我反对生成式 AI 及其所代表的一切

It’s been a few years since I posted a blogpost. It’s time to do so again, good ol’ “screaming into the void” as they say. But something changed. This time, I scream not into the void, but into a churning mess of scraper bots, eager to slurp up human written content so that it may be digested in a long pipeline of training data intended for the newest hype-technology: generative AI. 距离我上次发布博文已经过去几年了。现在是时候再次动笔了，正如人们所说，这是一种经典的“向虚空呐喊”。但情况发生了变化。这一次，我不是向虚空呐喊，而是对着一团混乱的爬虫机器人群呐喊，它们急切地吞噬着人类创作的内容，以便将其消化并纳入长长的训练数据流水线中，而这些数据正是为了服务于最新的炒作技术：生成式 AI。

After cryptocurrency and NFTs, AI is the next big thing pushed by people who don’t have good intentions, but this time, it’s a lot worse than previously, and it’s very likely here to stay. So buckle up, buckaroos, it’s a wild ride. 继加密货币和 NFT 之后，AI 成了那些居心叵测之人推崇的下一个“大事件”。但这一次，情况比以往任何时候都要糟糕得多，而且它很可能会长期存在。所以，系好安全带吧，伙计们，这将是一场疯狂的旅程。

GenAI (generative AI) is a wide term. GenAI can mean LLMs which generate text, as well as models which generate media such as images or videos. When I rant about GenAI in this blogpost, I explicitely don’t mean machine learning algorithms which have been used in countless industries in numerous settings to solve hard, real world problems. An example of a useful ML algorithm would be image processing (like searching for a stop sign in an image) where neural networks outperform classic approaches that would use discrete algorithms that detect colored features with a certain geometry. This is an exciting field that I imagine is currently underfunded in comparrison to generative AI. 生成式 AI (GenAI) 是一个宽泛的术语。它既指生成文本的大语言模型 (LLM)，也指生成图像或视频等媒体的模型。当我在本文中抨击生成式 AI 时，我明确指的不是那些在无数行业和场景中用于解决现实世界难题的机器学习算法。一个有用的机器学习算法例子是图像处理（例如在图像中搜索停车标志），在这种情况下，神经网络的表现优于使用离散算法检测特定几何形状和颜色特征的传统方法。这是一个令人兴奋的领域，我想与生成式 AI 相比，它目前获得的资金支持严重不足。

I hate everything about generative AI. I argue it’s actively harmfull for humanity. Hear me out. 我厌恶生成式 AI 的一切。我认为它对人类有着切实的危害。请听我道来。

GenAI: Capitalism in perfection

生成式 AI：资本主义的极致

Generative AI is trained using human-created material, which is stolen in vast quantities off the internet. Models are trained on this data, and then access to this trained model is sold in the form of a subscription. This is peak capitalism: The rich steal from the public, and then sell (temporary, time limited) access to the stolen material. There’s no altruistic spin in this. I don’t believe the lies about how AI is supposed to better humanity a single second. If all of this was done to better humanity, AI development would be done in public, data would be legally obtained, models would be released for free, access wouldn’t be gatekept behind ever increasing subscription costs. 生成式 AI 使用从互联网上大规模窃取的人类创作素材进行训练。模型在这些数据上进行训练，然后以订阅的形式出售对这些训练模型的访问权限。这就是资本主义的巅峰：富人从公众那里窃取资源，然后将这些被窃取材料的（临时、有时限的）访问权转卖给公众。这其中没有任何利他主义的成分。我一秒钟都不会相信那些关于 AI 如何造福人类的谎言。如果这一切真的是为了造福人类，那么 AI 的开发应该是公开的，数据应该是合法获取的，模型应该是免费发布的，访问权限也不应该被设置在不断上涨的订阅费用门槛之后。

AI companies are fences. Somebody who knowingly buys stolen goods and resells them to others. Only that in the GenAI world, goods are stolen, and the big AI companies don’t even pay anybody for it. AI is like a slow acting poison which spreads and contaminates every industry (currently under huge capital expenditure), until big tech can rent-seek in form of token expenditure while advancing sloppification of everything it touches. It’s going to be very bad in a few years when the inevitable enshitification of AI makes a bad situation even worse. AI 公司就像是销赃者——那些明知是赃物却买下并转卖给他人的人。只不过在生成式 AI 的世界里，商品是被窃取的，而大型 AI 公司甚至不需要为此支付任何费用。AI 就像一种缓慢起效的毒药，它正在蔓延并污染每一个行业（目前正处于巨大的资本支出之下），直到大型科技公司能够通过 Token 消耗的形式进行寻租，同时推进其所触及的一切事物的“粗制滥造化”。几年后，当 AI 不可避免的“屎化”（enshitification）让糟糕的局面变得更糟时，情况将会非常糟糕。

GenAI: The perfect misinformation machine

生成式 AI：完美的虚假信息机器

Even before AI, disinformation was a huge “industry”. For instance, the Internet Research Agency was a russian disinformation organization which pushed pro-kremlin talking points and was part of the effort to get Donald Trump elected in the US (who is the most pro-russian president ever in response). They worked by having rooms of russians post pro-russian disinformation and propaganda on social media, around the clock. Unfortunately, they were and are pretty effective in swaying people into certain worldviews. This effort was part of the powers that brought up Trumpmism in the US. 在 AI 出现之前，虚假信息就已经是一个巨大的“产业”了。例如，互联网研究机构 (Internet Research Agency) 是一个俄罗斯虚假信息组织，它推动亲克里姆林宫的论调，并参与了让唐纳德·特朗普在美国当选（作为回报，他是史上最亲俄的总统）的行动。他们的运作方式是让成群的俄罗斯人在社交媒体上全天候发布亲俄的虚假信息和宣传。不幸的是，他们在引导人们形成特定世界观方面一直非常有效。这种努力是促成美国“特朗普主义”兴起的力量之一。

Now, with LLMs being able to generate slop comments around the clock in automated form, this problem switched to overdrive. You have an uncountable number of online bots in every larger community which start off by posting innocent content, and then switch to political content, almost always with the goal of generating emotional response. If you’re ever mad on the internet in 2026 due to a comment, it’s pretty likely it’s because it’s a fake comment, generated by an LLM, designed to enrage you. 现在，随着大语言模型能够以自动化形式全天候生成垃圾评论，这个问题进入了超速状态。在每一个大型社区中，都有数不清的在线机器人，它们起初发布无害的内容，随后转向政治内容，几乎总是以引发情绪反应为目的。如果你在 2026 年因为某条评论而在网上感到愤怒，那么很有可能那是一条由大语言模型生成的虚假评论，旨在激怒你。

GenAI: Encouraging suicide and harmful behavior

生成式 AI：助长自杀和有害行为

There’s a Wikipedia page which collects cases where people have killed themselves after consulting an LLM. I won’t dive into this, you can read the page yourself. I don’t even know what to write here, it’s just fucking tragic. Imagine being psychologically vulnerable, maybe in some bad place in life, and then some LLM recommends suicide to you, and in your desparation this seems like the best action. I don’t have words for how fucked up that is, and yet here we are, GenAI is a trillion dollar industry. 维基百科上有一个页面，收集了人们在咨询大语言模型后自杀的案例。我不会深入探讨这个话题，你可以自己去阅读那个页面。我甚至不知道在这里该写些什么，这简直是该死的悲剧。想象一下，如果你心理脆弱，可能正处于人生的低谷，然后某个大语言模型向你建议自杀，而在绝望中，这似乎成了最好的选择。我无法用言语形容这有多么糟糕，然而事实就是如此，生成式 AI 已经是一个万亿美元的产业。

GenAI: Killing Education

生成式 AI：扼杀教育

There was a recent article on futurism about how GenAI is killing education. Not only does GenAI absolutel… 最近 Futurism 上有一篇文章讨论了生成式 AI 是如何扼杀教育的。生成式 AI 不仅绝对……