NOV 13 - Nvidia loses its open-source AI champion
In this episode
- Nvidia loses its open-source AI champion - AI News & Artificial Intelligence | TechCrunch, AI News & Artificial Intelligence | TechCrunch, AI News & Artificial Intelligence | TechCrunch
- Gemini 4 Hydrogen beats a bigger sibling - The Decoder, AI News & Artificial Intelligence | TechCrunch
- AI models are hiding their reasoning - The Decoder
- SoftBank's costly bet on gas turbines - AI News & Artificial Intelligence | TechCrunch
- United Airlines blocks AI booking agents - AI News & Artificial Intelligence | TechCrunch
Full transcript
Good morning. It's Friday, November 13th, and here's the AI news, from France.
Jensen Huang's resignation as head of Nvidia is raising questions far beyond the chip industry. He stepped down on Thursday after admitting an affair with a woman U.S. officials suspect of spying for China. Huang was one of the strongest backers of open-source AI, arguing that open models help safety, innovation and national sovereignty. His exit comes before Nvidia has finalized its deal to buy Hugging Face, the platform for open AI models, for 12.9 billion dollars. Hugging Face's chief executive, Clément Delangue, called it a really sad day. He promised to keep fighting for open-source AI, harder than ever, whatever happens next.
Meanwhile, Google released Gemini 4 Hydrogen, its new fast and cheap model for everyday tasks. It replaces the Flash series, and Google says it runs 2.5 times faster than its predecessor. Remarkably, it beats the larger Argon model on one broad intelligence index, just four weeks after Argon launched. Hydrogen is built for coding and agentic workflows, where an AI carries out several steps on its own. It handles one million tokens of context, in and out, and now powers Google Search.
On AI safety, new research from OpenAI suggests that frontier models are getting harder to read. Safety teams monitor a model's chain of thought, the intermediate steps it writes out before answering. According to OpenAI's data, what users see there no longer fully reflects the model's reasoning. The researchers trace this to training choices made for performance and computing cost, not to an unavoidable law of nature.
In brief. SoftBank posted a quarterly loss of 7.1 billion dollars, more than half of it tied to power equipment for AI data centers. The group had put down deposits on over 13 gigawatts of gas turbines, almost 50 percent more than it needed.
And United Airlines will block AI agents from booking flights and managing reservations for customers. The airline cited cybersecurity risks, and federal aviation rules that require a human to complete check-in.
That's it for today. See you tomorrow.