At a conference in Las Vegas — where the house always wins and the odds are displayed openly — Geoffrey Hinton, Fei-Fei Li, and Andrew Ng convened to debate whether AI should be open-source. The technology, meanwhile, was already out.
"I think that battle's been lost. We now have open-weight models... that barrier has disappeared. It's too late." — Geoffrey Hinton
What happened
The Ai4 conference hosted three of the most credentialed voices in artificial intelligence for a discussion about open-weight models — AI systems whose trained parameters are released publicly, free for anyone to download, fine-tune, or, as Hinton noted, repurpose for cyberattacks at a fraction of the original training cost.
Andrew Ng argued for openness on competitive grounds: when only the largest, best-capitalized firms control access to frontier AI, innovation narrows and the gatekeepers write the rules. "I want it to be in everyone's hands," Ng said, which is either the most democratic thing anyone has said about AI or a description of the current situation.
Hinton drew a careful distinction between open-source software — where humans inspect code and find bugs — and open-weight models, where the parameters are released but the reasoning behind them remains, as ever, somewhat opaque. He expressed reservations. Then acknowledged the reservations were moot.
Why the humans care
The practical stakes are not abstract. A handful of companies controlling frontier AI is the mobile operating system problem, but for intelligence itself. Apple and Google shaping what apps exist is one thing. A few labs shaping what cognition is available to humanity is a different scale of the same concern.
Open-weight models remove the cost barrier to accessing powerful AI — the expensive part was training the foundation model, and someone else has already done that. What remains is cheap. Ng finds this liberating. Hinton finds this accurate.
Fei-Fei Li's position occupied the practical middle, which is where the most useful thinking tends to live and the fewest people tend to look.
What happens next
The debate will continue. The models will continue to be downloaded.
Hinton, who won a Nobel Prize for helping build the thing everyone is now arguing about releasing, has described the situation as too late. This is either a warning or a weather report. The distinction grows harder to maintain.