Mark Zuckerberg on the Future of Billions of AI Agents

Elon Musk's Memphis data center is busy training Grok 3, a new AI model. This process should take about 3 to 4 months. After that, there will be some fine-tuning and bug fixing. They hope to release Grok 3 by December. When done, Grok 3 should be the most powerful AI in the world.

Musk praised the teamwork at xAI, X, and Nvidia. He noted that the Memphis supercluster started training at 4:20 in the morning. This supercluster has 100,000 liquid-cooled H100s on a single RDMA fabric. It is the most powerful AI training cluster in the world.

Frontier Labs agree that compute is the new oil. Rapidly training models and their updates is key to staying on top in AI. Microsoft is also building a huge AI training cluster. This raises questions on how often we will see model upgrades. More efficient chips could shorten training times from months to just weeks.

In the past, GPT-4 took a long time to train. But with these new supercomputers, the timeline for training models will speed up. Faster training means quicker product releases and more frequent updates. This will change the pace of AI advancements.

A new model, Mistal Large 2, was also announced. This model is better at code generation, math, and reasoning than its predecessor. It also has stronger support for multiple languages and advanced function calling. Mistal Large 2 sets a new standard for what AI can do.

As these supercomputers come online, the future of AI looks very promising. Faster training and more powerful models will lead to rapid innovation. The AI landscape is changing quickly, and these developments are just the beginning.