Garry Tan wants US open-weight AI labs to 'distill' frontier models, too
402 points
• 1 day ago
• Article
Link
Y Combinator 的 CEO Garry Tan 反对监管机构打击 AI 蒸馏（distillation）的做法。所谓蒸馏，是指开发者通过对一个 AI 模型的提示，诱导其暴露信息处理方式，再利用这些输出训练另一个模型。
尽管包括 Anthropic 的 CEO Dario Amodei 在内的一些行业领袖呼吁政府介入，以防他们所称的非法蒸馏攻击，Tan 认为不应实施这样的监管。相反，他主张美国应当建立并鼓励自己的蒸馏体系，支持规模较小的美国开放权重（open-weight）AI 实验室对前沿模型采用这些训练方法，从而形成更有竞争力的格局。通过此举，美国可以培育起强大的开放权重替代生态，以挑战大型专有 AI 提供商的主导地位。
Tan 强调，他并不鼓励使用被盗凭证或欺诈手段访问模型；他认为前沿 AI 实验室无权限制客户如何利用通过 API 交互获得的信息。他还指出，用知识产权为由反对蒸馏具有讽刺意味——这些专有机构在训练自身系统时，往往未经许可就吸收了大量受版权保护的人类知识。
在 Tan 看来，该行业面临的最大危险是权力被某个单一巨头垄断。他警告，如果一家拥有雄厚资本和顶尖研究人才的专有供应商被允许一枝独秀，市场将走向对创新和公众访问极为不利的局面。支持开放权重模型，是他希望确保人工智能的强大能力保持可访问性与平衡、而不是被限制性服务条款锁死的一条路径。
Y Combinator CEO Garry Tan is pushing back against the idea that regulators should crack down on AI distillation. Distillation is a technique where developers use one AI model to train another by prompting it to reveal how it processes information. While some industry leaders, including Anthropic CEO Dario Amodei, have called for government intervention to prevent what they describe as illicit distillation attacks, Tan believes there should be no such regulation.
In fact, Tan argues that the United States should embrace a distillation regime of its own. He suggests that smaller, American open-weight AI labs should be encouraged to use these training techniques on frontier models, effectively creating a more competitive landscape. By doing so, the U.S. could foster a robust ecosystem of open-weight alternatives to challenge the dominance of massive, proprietary AI providers.
Tan emphasizes that he is not advocating for the use of stolen credentials or fraudulent means to access these models. Instead, he maintains that it is an overreach for frontier AI labs to dictate how their customers utilize the information gained through API interactions. He also points out the irony in proprietary labs claiming intellectual property concerns, noting that these same organizations often ingested vast amounts of copyrighted human knowledge without permission to train their own systems in the first place.
From Tan's perspective, the primary danger facing the industry is the concentration of power in the hands of a single, monolithic company. He warns that if a proprietary provider with superior capital and research talent is allowed to run away with the market, it would create a nightmare scenario for innovation and public access. By championing open-weight models, he hopes to ensure that the immense power of artificial intelligence remains accessible and balanced, rather than locked behind restrictive terms of service.
223 comments • Comments Link
- AI 的基本伦理存在争议。有人认为 frontier models 主要建立在对受版权保护作品的未经授权"掠夺式开采"之上，因此开发者关于限制这些模型使用的道德主张显得站不住脚。
- 对 intellectual property 安全的担忧很高。许多用户对 "Zero Data Retention" (ZDR) 政策持怀疑态度，认为这不过是来自曾有绕过 copyright 倾向的机构的、难以执行的口头承诺。
- 把公司称作 "labs" 的行业做法越来越受批评：这被视为一种营销策略，用来塑造科学与亲社会发现的形象，而实际上是通过大量挖掘公共数据来生产 proprietary 产品。
- Model distillation（用一个模型的输出去训练另一个模型）被一些人视作必要的竞争手段和一种 "fair use"，尤其是在原始模型本身建立在未经授权使用人类集体知识的基础之上时。
- Regulatory capture 是一项重大关切：人们担心现有的 frontier labs 正在借助 "safety" 叙事寻求政府干预，以阻止竞争对手获得他们首创的技术。
- Open-weights models 的支持者认为，让 AI 获取更加民主化对长期创新至关重要，这可防止少数实体垄断主要的 intellectual production 引擎。
- 人们对当前 AI 商业模式的可持续性仍持怀疑：高额训练成本和模型的商品化表明现行融资策略在结构上可能存在问题。
- 对 closed-weight models 的不信任正在上升。有声音警告，隐藏的偏见、潜在后门或强势实体的意识形态篡改，可能使 open-weights 比 proprietary 服务的 "mystery-box" 特性更受欢迎。
- AI 开发的经济模式受到质疑：如果公司的商业模式依赖于从公共资源中提取价值却不作公平补偿，那么它们就没有权利要求保证的投资回报。
- 有人提议将领先的 AI laboratories 国有化，或强制为 distillation 提供 "fair-price" 的获取渠道，作为潜在保障，以确保 AI 能力成为公共产品而非被锁定的 proprietary 资产。
这种讨论反映出 pioneering AI labs 与更广泛开发者社区之间根深蒂固的紧张关系，主要源于对 intellectual property 被窃取和排他性市场策略的认知。公众明显更支持 open-weights models，视其为抵御垄断控制和 black box 系统不透明性的一种手段。尽管部分利益相关者担心持续对训练数据进行 distillation 会损害 AI 研究的经济可行性，越来越多的人认为，当前建立在不透明数据抓取和 regulatory gatekeeping 之上的路径在道德和运营层面上都是不可持续的。 • The fundamental ethics of AI are contested, with claims that frontier models exist primarily by "strip-mining" copyrighted works without authorization, thereby invalidating any moral claim by developers to restrict how those models are used.
• Concerns regarding intellectual property security are high, as many users remain skeptical of "Zero Data Retention" (ZDR) policies, viewing them as unenforceable "pinky-promises" from entities that have already demonstrated a willingness to bypass copyright norms.
• The industry practice of labeling firms as "labs" is increasingly criticized as a marketing tactic designed to project an image of scientific, pro-social discovery, while the reality involves consuming vast amounts of public data to create proprietary products.
• Model distillation—using the output of one model to train another—is viewed by some as a necessary form of competition and a form of "fair use," especially since the original models were built upon unauthorized usage of collective human knowledge.
• Regulatory capture is a significant concern, with fears that current frontier labs are leveraging "safety" narratives to solicit government intervention that prevents competitors from accessing the technologies they pioneered.
• Proponents of open-weights models argue that democratizing access to AI is essential for long-term innovation, preventing a monopolistic future where only a few entities control the primary engines of intellectual production.
• Skepticism persists regarding the sustainability of current AI business models, as high training costs and the commoditization of models suggest that current funding strategies may be structurally flawed.
• Distrust of closed-weight models is growing, with warnings that hidden biases, potential backdoors, or ideological tampering by powerful entities could make "open weights" preferable to the "mystery-box" nature of proprietary services.
• The economic model of AI development is being challenged by the assertion that companies are not entitled to a guaranteed return on investment if their underlying business practices relied on extracting value from the public commons without fair compensation.
• Nationalizing leading AI laboratories or mandating "fair-price" access for distillation are proposed as potential safeguards to ensure that AI capabilities function as a public good rather than a locked-down proprietary asset.
The discourse reflects a deep-seated tension between the pioneering AI labs and the broader developer community, fueled by perceptions of intellectual property theft and exclusionary market tactics. There is a palpable shift toward favoring open-weights models, which are seen as a hedge against both the potential for monopolistic control and the opacity of "black box" systems. While some stakeholders worry about the economic viability of AI research if training data is continuously distilled, a growing consensus suggests that the current path—built on opaque data scraping and regulatory gatekeeping—is ethically and operationally unsustainable.