Thom_Wolf has a two-word reply to calls to keep building in the open: “We will.” In a November 13 post on X, he said “plenty inside NVIDIA” had been messaging his team with that encouragement, and argued that open AI models don’t depend on a single champion. Source
His post quotes Jensen Huang—but the endorsement is from July 24, not a fresh NVIDIA announcement. In that earlier post, Huang shared a letter he said NVIDIA had signed and wrote: “The world needs both frontier closed models and frontier open models.” SourceSource
The linked letter, Open Weights and American AI Leadership, describes open-weight models as models people can download, inspect, modify and run on their own infrastructure. Its signatories, including NVIDIA and Hugging Face, argue that this gives organizations more control and lets them adapt advanced models without training one from scratch or depending on a single provider. Source
The letter also acknowledges that openness doesn’t come with an undo button. Once weights are released, it says, they’re beyond the original developer’s control, and modified versions are difficult to trace or reverse. It nevertheless argues that broader access lets more teams examine model behavior, identify vulnerabilities and develop safeguards. Source
Wolf also said Huang spent part of a meeting a couple of weeks earlier discussing “our little $400 open-source microrobot.” His questions, according to Wolf, covered heat, chips, computing performance and power draw. Even a tiny robot gets a thermal-management subplot. Source
The post announces no specific release, deadline or NVIDIA partnership. Source
Comments