在拉斯维加斯举办的Ai4会议上，三位顶级AI研究人员就开源AI模型的安全性与监管问题展开讨论[1]。诺贝尔奖得主Geoffrey Hinton、World Labs CEO Fei-Fei Li和Coursera联合创始人Andrew Ng虽在具体策略上存在分歧，但均主张保持AI的开放性，以防止少数大公司垄断技术进展[1]。
Hinton坦承开源模型已成现实且难以改变[1]。他曾反对开放模型权重，理由是"这使得人们可以轻易获取训练成本极高的大型基础模型，然后花费少得多的成本将其训练用于网络攻击等不当用途"[1]。然而他认识到"这场战斗已经输了"，因为"开源模型的出现消除了训练成本这一障碍"[1]。
相比之下，Ng强调开源AI的竞争优势和软实力意义，主张"不应该有守门人，那会限制我们所有人获取AI的方式"[1]。他表示"如果要给出一个建议，那就是推动开放，因为AI是一项了不起的技术，我希望它掌握在每个人手中"[1]。
Fei-Fei Li则主张在极端开放与完全封闭之间寻求平衡[1]。她指出"把这个问题简化为完全开放或完全封闭的二元对立是非常危险的，在复杂软件系统和科学系统中，需要更细致的考量"[1]。
三人共同认为某种程度的监管对引导AI正确发展是必要的[1]。Hinton表示"我们需要朝着帮助人类的方向发展AI，而监管将帮助我们实现这一点"[1]。
At the Ai4 conference in Las Vegas, three leading artificial intelligence researchers engaged in a substantive debate over the future of open-source AI models and their governance. Nobel Prize winner Geoffrey Hinton, World Labs CEO Fei-Fei Li, and Coursera co-founder Andrew Ng each articulated distinct positions on balancing openness with safety, yet converged on the principle that AI development should remain accessible rather than concentrated among a handful of major corporations.[1]
Hinton acknowledged a fundamental shift in the landscape of AI development, stating that "open-weight models" have become an irreversible reality.[1] He explained his previous opposition to open releases by noting that "it makes it so easy for people to take these big foundation models, which are very expensive to train, and for much less money train them to do bad things like cyber attacks," but conceded that "that battle's been lost."[1] The elimination of high training costs as a barrier to access, Hinton argued, has fundamentally altered the calculus around containment.[1]
Ng took a more emphatic stance on accessibility, asserting "I don't want there to be gatekeepers. That limits how all of us can access AI."[1] He advocated for widespread distribution of AI capabilities, declaring "if I were to try to give one prescription, it would be to promote openness because AI is amazing technology and I want it to be in everyone's hands."[1]
Li offered a middle ground, warning against framing the issue as binary. She stated: "It's very dangerous to make this a dichotomy between complete openness all the way to complete closedness. In complex software systems as well as scientific systems it's much more nuanced."[1]
Despite their disagreements on strategy, the three researchers found common ground on governance. Hinton emphasized that "regulation will help us" achieve the goal of developing "AI in a direction that helps people," suggesting that some form of oversight is necessary to guide AI toward beneficial outcomes.[1]