Techno
Training AI models might not need enormous data centres

Heyonce again, The world’s richest men competed for yachts, planes and private islands. Now, the favorite competition for measuring volume is clusters. Just 18 months ago, it openedAmnesty International training gpt-4, the large language model developed at that time (Master’s), on a network of about 25,000 advanced graphics processing units (GPUs) Made by Nvidia. Now, Elon Musk and Mark Zuckerberg, the heads of X and Meta respectively, are waving their chips in the air: Musk says he has $100,000. GPUIn one data center and plans to buy 200,000. Mr. Zuckerberg says he will get $350,000.