Foreign Media: Three AI Pioneers Oppose Comprehensive Tightening of Open Models

As the debate over AI safety intensifies, foreign media reports that Geoffrey Hinton, Fei-Fei Li, and Andrew Ng recently discussed the issue of open models at the Ai4 conference in Las Vegas. While their positions are not entirely aligned, all three oppose the dominance of a few large tech companies in the pace of AI development.

Opposing a Few Companies Controlling Access

The article states that the three researchers share a common concern: if AI is controlled by a few companies, the space for innovation may be compressed. Ng explicitly stated that he does not want to see "gatekeepers" in the AI field, as this would limit how developers, businesses, and the public can access AI.

He believes that large companies will naturally protect their own advantages, which could influence the formation of industry rules. The result may be that only the companies with the most funding and computing power can continue to build the most advanced systems. Ng suggests maintaining a coexistence of multiple model providers, allowing models and companies to compete in the market rather than letting a few platforms dominate.

Hinton Distinguishes Open Source Code from Open Weights

However, Hinton does not equate "open source software" with "open weight models." According to him, open source software makes the code public, allowing external parties to check for vulnerabilities and propose modifications; open weights, on the other hand, directly hand over the parameters of a completed large model to the outside world, significantly lowering the barrier for secondary development.

He stated that he previously opposed open weights because it would allow some individuals to repurpose foundational models into tools for cyber attacks at a lower cost. However, he also acknowledges that this debate has become difficult to reverse in reality. With the continuous emergence of open weight models, the barriers that training costs once constituted have been significantly weakened.

Nevertheless, Hinton does not advocate for halting AI development. The article notes that he still believes AI will enhance productivity and improve fields such as education and healthcare, but we cannot ignore the potential risks. He also mentioned that it is unfair to label all those who express concerns about AI risks as "panic creators."

Li Fei-Fei Advocates for Layered Openness

In contrast to Ng's emphasis on open competition and Hinton's focus on safety risks, Li Fei-Fei opposes simplifying the discussion into two choices: "completely open" or "completely closed." She believes that complex software and research systems inherently exist at different levels, and different stages can adopt varying degrees of openness.

Using nuclear physics as an example, she stated that academic papers can be publicly published, but uranium materials are strictly regulated, and experimental phases fall somewhere in between. Based on this, she argues that AI can adopt a similar approach: maintaining a certain level of openness in research discoveries, education, and international cooperation while allowing companies to retain closed-source systems and business models.

The article also mentions that Li Fei-Fei cites the Human Genome Project to illustrate that foundational knowledge platforms formed through collaboration between public institutions and the private sector can simultaneously support research advancement, corporate profit, and social application. She believes that AI infrastructure should also develop in this direction rather than getting caught up in an either-or debate.

All Three Recognize the Need for Regulation

Although the three have different judgments about the boundaries of open models, their attitudes toward regulation are similar. The article states that they all believe AI still requires a certain degree of regulatory constraints to develop in a direction more beneficial to society.

Hinton stated that the development of AI should not be left entirely to entrepreneurs to decide. The current debate surrounding open models is no longer just a technical route dispute; it also relates to the distribution of industry power, international competition, and the direction of AI safety governance.