On my own. Elon Musk said on X Tuesday that xAI, the company he founded to “destroy OpenAI,” is building its own data center from scratch. As part of that push, it has purchased 100,000 Nvidia H100 GPUs, making it what Musk called “the most powerful training cluster in the world by a large margin.” The move is classic Musk. The company launched less than a year ago and already has a data center 10 times the size of the one used to create GPT-4. If each GPU costs somewhere around $30,000, we’re talking about a $3 billion bet (not including energy costs) that the company can create the most powerful AI model in the world. Musk’s company had worked with Oracle on previous training runs to create the chatbot Grok and the upcoming Grok 2. “The reason we decided to do the 100k H100 and next major system internally was that our fundamental competitiveness depends on being faster than any other AI company. This is the only way to catch up,” he wrote. This is significant because it likely puts xAI in a rarefied group of foundation model companies like Google, OpenAI, Anthropic and Meta. OpenAI is building a cluster with 100,000 Nvidia BG200 GPUs, a more powerful successor to the H100, The Information reported. But that won’t open until the second quarter of next year. Meanwhile, Andreessen Horowitz is reportedly eyeing up to 20,000 GPUs in its arsenal to help its portfolio companies. Mario Anzuoni/File Photo/ReutersIt’s complicated. The New York Times’ relationship with OpenAI isn’t as straightforward as it seems. It sued the startup last year, accusing OpenAI of copyright violations for ripping off its articles to train the company’s large language model. But while they fight each other in court, the NYT is also testing OpenAI’s technology behind closed doors. The Intercept found that the publisher had built a tool, powered by the startup’s technology, to automatically generate headlines and edit copy to make sure the writing adheres to its style guide. A representative from the NYT said that the software was an experiment and wasn’t used by the newsroom. Several news outlets have begun using AI, but these applications haven’t been well received. The media industry is suspicious of AI, and are concerned that the technology could end up replacing journalists. The NYT’s experimental tool, for example, performs the duties of an editor. As AI improves, it’s probably only a matter of time before top publishers like the NYT start using it in their newsrooms. The Washington Post, for example, just launched Climate Answers, an AI chatbot that answers questions about the environment based on its own reporting. Also, in other OpenAI news: Microsoft has given up its observer board seat at the firm amid antitrust concerns. Regulators in the US and UK are investigating their close relationship, which involves Microsoft supplying compute and OpenAI providing AI products. — Katyanna Quach |