At the Ai4 conference in Las Vegas this August, three of the most-cited names in modern AI - Nobel laureate Geoffrey Hinton, Stanford’s Fei-Fei Li, and Coursera co-founder Andrew Ng - shared one stage and walked off with three separate positions on whether open-weight AI should be celebrated, constrained, or split into layers.
The 2026 open-source-versus-regulation debate is not short on voices. What made this panel unusual was that all three landed in one venue, in front of an audience of 12,000 attendees, and got pressed on the same questions. The honest read of the transcript is that they could not even agree on what to call the debate. The broader fight over who gets to write the rules has been tracked in our AI Regulation Tracker coverage, and Ai4 was a clean snapshot of how unsettled the field still is.
Where they stood
According to TechCrunch’s 12 August coverage of the panel, the discussion opened with a question none of the three answered the same way. Hinton drew a sharp line between open-source code and open-weight models. “[Open source] means you show people the code, and lots of people look at the lines of code and say, ‘Oh, there’s a bug,’” he said. “Open weights means you train a big model and then you give people the weights. That’s very different.” He told the audience he had originally been against open weights because they make it easy for bad actors to fine-tune large models for cyber attacks. Then he conceded the fight was already over: “I think that battle’s been lost. We now have open-weight models, so the barrier to lots of people getting these big models, which was the cost of training foundation models, that barrier has disappeared. It’s too late.” The next words were the ones that mattered. “What we want to do is develop AI in a direction that helps people,” he said, “and regulation will help us do that. You can’t leave it to people like Elon Musk and Mark Zuckerberg to decide how AI should be done.”
Ng landed in the opposite direction. His argument, per the same TechCrunch write-up, was that US fear-mongering about AI risk is the single biggest gift American open-source labs could give their Chinese competitors. “But my worry is because of all the lobbying in the U.S. and the fear-mongering, building open source AI in America is struggling to compete with open-weight models coming out of China,” he said. His one-line policy prescription was direct: “promote openness, because AI is amazing technology and I want it to be in everyone’s hands.” He did not want gatekeepers. He framed open-source AI as soft power, and warned that if China figures out a fundamentally more cost-efficient way to build AI, “then things that are more cost-efficient have a fundamental business adoption advantage.” That dynamic is exactly the one we covered when DeepSeek shut Nvidia out of its V4 model and gave Huawei a weeks-long head start.
Li rejected the binary entirely. According to Forbes’ 6 August recap and confirmed by TechCrunch’s same-day coverage of the Ai4 panel, she called the framing “a false debate” and proposed a layered model. “It’s very dangerous to make this a dichotomy between complete openness all the way to complete closedness,” she said. Her analogy, covered by R&D World Online the same day, was nuclear physics: scientific papers stay open, fissile materials stay regulated, and the in-between work happens in licensed labs. “In complex software systems, and in scientific systems, the picture is much more nuanced than the words open versus closed,” Li said. “Let’s bring science, not science fiction, back to the AI debate. The public needs education. It needs respect. It needs good, healthy scientific communication. AI is a tool.”
The narrow overlap
The closest thing to consensus was a single word: regulation. Hinton endorsed it. Forbes reports that he framed the choice bluntly: “Developing AI is like the accelerator. Regulation is like the steering wheel.” Li endorsed sector-specific rules, the way drugs, financial products, and aviation are regulated. Ng did not directly endorse regulation but did not reject the framing either. The pre-keynote announcement from the conference positioned the three as a combined “case for staying open” - but the panel transcript makes clear that what stayed open, and at what layer, is exactly the part nobody could settle.
A second small overlap was on jobs, though even there the three did not converge. Forbes and R&D World each have Hinton framing displacement through the backhoe analogy: “My judgment is that many jobs will go the way of ditch diggers when backhoes came along.” Per R&D World, he added that paralegal work is “more or less disappearing because AI can do that job better than people,” and that any job built around routine intellectual labor is now at risk. Li pushed back on the replacement framing. “When people put the words AI and jobs together,” she said per R&D World Online, “the implicit word between them is ‘replace.’” Her policy pitch was a “soft landing” via retraining and sector-specific rules. Ng cited a CEO survey showing 1.4% of laid-off workers were actually replaced by AI and called the broader job-impact narrative “blown out of proportion.”
What This Means
The question worth keeping is not which of the three is right. It is who actually has to build the compromise. The audience at Ai4 - 12,000 attendees and more than 1,000 speakers, per the conference’s own announcement - is largely enterprise software buyers, AI vendor staff, and research-lab employees. Those are the people whose quarterly product roadmaps decide whether the next round of models ships with weights attached or behind an API. Li’s layered model only becomes concrete if it lands inside a shipping decision: a license, a usage clause, a tier of weights withheld. Until then, the field sits at Hinton’s “too late” and waits for a regulatory body that does not yet exist.
The Bottom Line
Three of the most-cited names in AI could not agree on what to call the open-source debate, let alone how to settle it. The smallest common ground was that regulation belongs in the conversation. Whether that is enough to build anything is a question the next round of lab founders will answer with code, not keynotes.