OpenAI 与微软早知自己正在为互联网开启“末日循环”

近期解封的法庭文件，在_纽约时报_起诉 OpenAI 和微软案中对两家公司颇为不利。这些公司自己的文件警告称，它们正在开启一个会损害互联网的“末日循环”；文件将其为训练模型而抓取数据的行为形容为“人类历史上最大规模的劳动窃取”，并称这让“合理使用这一理念彻底沦为笑话”。

文件中许多最引人注目的引述来自微软应用科学总监 Brent Hecht。不过，微软试图与 Hecht 的论断保持距离。微软发言人 Alex Haurek 向 The Verge 表示：“这些评论反映的是一名员工的个人观点，不构成法律分析，也不代表公司立场。”

在另一份法庭文件中，微软 AI 数据战略与运营总经理 Jordan Usdan 将 Hecht 的角色描述为具有对抗性。他表示，Hecht“对于 AI 数据生态系统应如何运作持有不同的、学术性的和前瞻性的观点；微软聘用他，是为了带来非对称的、未来主义的和学术性的视角……他也并非代表微软就其有关 AI 对内容创作者潜在影响的理论观点发言的人。”

但无论微软是否愿意认领这些评论，显然，这已经成为现实。Google Zero 真实存在！AI 正在吞噬互联网！

《纽约时报》的文件中还有大量来自不同人士的惊人表述，包括 Satya Nadella、Sam Altman 以及其他 OpenAI 员工。以下是这份 92 页文件中的部分重点。

引言引用了 Hecht 和 OpenAI ChatGPT 负责人（推测为 Nick Turley）的说法，似乎表明两家公司知道自己会对像_纽约时报_这样的出版商构成“生存性威胁”。Hecht 称 ChatGPT 和 Copilot 对数据的收集是“人类历史上最大规模的劳动窃取”，并称微软的辩护让“‘合理使用’这一理念彻底沦为笑话”。

Satya Nadella 承认，聊天机器人实际上已经取代了搜索，并消除了人们直接前往信息源头的必要。但或许更具杀伤力的是一份微软内部文件，其中写道：“我们的 AI 内容战略已开启一个会同时损害模型表现和整个网络的‘末日循环’：一个终端产品威胁其关键供应商经济基础的情况极不寻常，但这正是我们围绕‘内容供应链’为 LLM 业务创造的局面。”

那甚至都不算一个真实数字

不要被 OpenAI 或微软关于利他意图的说法蒙蔽。OpenAI 联合创始人 Greg Brockman 更感兴趣的是，他可能通过商业 AI 赚到“数不清的巨额资金”。

尽管后文引用 Nadella 的话称，“任何付费墙后的内容都应该获得授权”，一名 OpenAI 代表承认，他“不知道”公司是否曾努力检测或从训练数据中移除付费墙内容。

“极其擅长复述”

从内部情况看，OpenAI 似乎非常清楚 ChatGPT 倾向于“逐字”复现受版权保护的材料。尽管公司承认，“防止记忆化”对于“尽量减少版权侵权”很重要，但员工承认 GPT-4“记住了大量数据，因此会极其擅长复述”。

该文件随后列举了多个例子：面对查询时，ChatGPT 直接输出了来自_Times_、Mercury News、The Denver Post、LifeHacker 和 Eurogamer 文章的大段原文。

“把他们的所有作品都吸走”

微软知道其对互联网进行全面抓取会被如何看待，并承认“几乎没有人希望他们[原文如此]创作的内容以这种方式被使用，他们也没有因其使用而获得报酬”。

“替代人类劳动”

OpenAI 政策总监 Jack Clark 已看出事态发展方向。他表示，公司正在“创建替代那些定义社会‘文化’之人的劳动的系统”。内部文件将 ChatGPT 描述为“现代报摊”。后文引述 OpenAI 的 Nick Turley 称，一旦从聊天机器人获得答案，“就没有充分理由”再点击通往原始来源的链接。

摧毁自己的供应链

文件引用微软的承认称，“LLM 是一种会摧毁自身供应链的产品”，因为在许多情况下，它会替代其自身的训练数据。

OpenAI 知道自己正在扼杀引流流量

OpenAI 自己的媒体和经济专家将_Times_等网站推荐流量的下降，直接归因于 Google AI Overviews 等 AI 摘要功能。他们推测，来自搜索的推荐流量可能已下降多达 60%。

微软发言人 Haurek 警告称：“Satya 的证词与微软在本案中的立场完全一致。他谈到的是人们寻找和消费信息方式正在发生的广泛原则和变化。不能将这些观察与法庭面前的版权问题结论混为一谈；微软已在其提交的文件中回应了这些问题。”

但根据这份新近解封的文件，似乎相当明确的是，微软和 OpenAI 都知道自己将对出版业、该行业雇用的“数百万人”，以及进而对自身产品造成不可弥补的伤害，但为了赚取“数不清的巨额资金”仍执意推进——哪怕这意味着走向末日循环。

Recently unsealed court documents in the New York Times ’ case against OpenAI and Microsoft are pretty damning. The companies’ own documentation warned that it was starting a “doom loop” that would damage the web, characterized its scraping of data to train its models as the “largest theft of labor in human history,” and that it made a “complete mockery of the idea of fair use.”

Many of the most eye-catching quotes from the document come from Microsoft’s Director of Applied Science, Brent Hecht. Though, the company has tried to distance itself from Hecht’s assertions. Microsoft spokesperson Alex Haurek told The Verge that “These comments reflect one employee’s individual perspective, are not a legal analysis, and do not represent the company’s views.”

In a separate court filing, Jordan Usdan, GM for Data Strategy and Ops at Microsoft AI, characterized Hecht’s role as adversarial. He said that Hecht “holds divergent, academic, and forward-looking views about how data ecosystems for AI should operate and is employed at Microsoft to bring asymmetrical, futuristic, and academic points of view … nor is he someone who speaks for Microsoft specifically as to his theoretical views on AI’s potential effect on content creators.”

But whether or not Microsoft wants to own these comments, it’s clear that this came true. Google Zero is real! AI is eating the web!

There are plenty more wild statements in NYT’s filing from a variety of figures, including Satya Nadella, Sam Altman, and other OpenAI employees. Here are some highlights from the 92 page document.

The introduction quotes Hecht and OpenAI’s Head of ChatGPT (presumably Nick Turley) in a way that seems to show the companies knew they posed an “existential threat” to publishers like the New York Times. Hecht calls ChatGPT and Copilot’s harvesting of data the “largest theft of labor in human history” and says that Microsoft’s defense makes a “complete mockery of the idea of ‘fair use.’”

Satya Nadella admits that chatbots have basically replaced search and removed the need to go straight to the source for info. But perhaps more damning is an internal Microsoft document that says, “Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time: It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business with respect to its ‘content supply chain.’”

That’s not even a real number

Don’t be fooled by OpenAI or Microsoft’s claims of altruistic intent. OpenAI cofounder Greg Brockman is more interested in the “gazillions” of dollars it he could potentially make through commercial AI.

Despite Nadella later being quoted as saying, “anything that is paywalled should be licensed,” An OpenAI representative admitted that he was “unaware” of any effort to detect or remove paywalled content from training data.

“Insanely good at regurgitation”

Internally, it seems that OpenAI was well aware of ChatGPT’s tendency to simply reproduce copyrighted material “verbatim.” Even though it acknowledged that the “prevention of memorization” was important to “minimize copyright violations,” employees admitted that GPT-4 “memorized a ton of data and therefore will be insanely good at regurgitation.”

The filing then goes on to cite several examples of ChatGPT outputting long strings of copy straight from articles in the Times , Mercury News , The Denver Post , LifeHacker , and Eurogamer in response to queries.

“‘Hoovering up’ all their work”

Microsoft knew how its wholesale scraping of the internet would be perceived and admitted that “almost no one intended for they [sic] content they created to be used in this fashion, nor are they compensated for its use.”

A “substitute for the labor of people”

OpenAI Policy Director Jack Clark saw the writing on the wall, saying that it was “creating systems that substitute for the labor of the people that define the ‘culture’ of society.” Internal documents described ChatGPT as “the modern newsstand.” OpenAI’s Nick Turley is later quoted as saying that once you get an answer from its chatbot, there is “no good reason to click” on a link to the source.

Destroying their own supply chain

Microsoft is quoted as admitting that “LLMs are a product that destroys its own supply chain” because it’s a substitute for its own training data in many cases.

OpenAI knows its killing referral traffic

OpenAI’s own media and economic experts attributed the drop in referral traffic for sites like the Times directly to AI summaries like Google’s AI Overviews. They’ve speculated that search referrals may be down as much as 60 percent.

Microsoft spokesperson Haurek cautioned that “Satya’s testimony and Microsoft’s position in this case are perfectly consistent. He spoke to broad principles and changes underway in how people find and consume information. Those observations should not be confused with conclusions about copyright questions before the Court, which Microsoft addresses in its filings.”

But it seems pretty clear based on this newly unsealed document that both Microsoft and OpenAI knew they were going to irreparably harm the publishing industry, the “millions of people” it employs, and, by extension, damage their own product, but carried forward anyway in pursuit of “gazillions” of dollars — doom loop be damned.

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Terrence O'Brien

• AI

• Google

• Law

• Microsoft

• OpenAI

• Policy

• Tech