On July 22, Elon Musk announced on X that the Memphis supercomputer cluster, built by xAI, X, NVIDIA, and other supporting teams, began training at 4:20 AM local time.
The cluster consists of 100,000 liquid-cooled H100 chips, making it the world's most powerful supercomputer cluster. In an interview with psychologist Jordan Peterson, Musk revealed that it took only 19 days from hardware installation to training commencement.
Musk also shared that Grok 2's training is complete and will be released next month. Grok 3 is currently being trained in the Memphis supercomputer center and is expected to be released by the end of the year, potentially becoming the most powerful model on the market.
However, Musk expressed growing concerns about AI safety. He believes AI must be based on "pro-human" principles, contrasting with other companies' goals of creating digital superintelligence far smarter than humans.
Regarding Grok's current performance, Musk admitted that the basic model on X is not very strong, being an order of magnitude weaker than ChatGPT. He emphasized xAI's rapid progress, noting that they built a large training center in just 19 days.
The Memphis "supercomputer factory" features 100,000 liquid-cooled H100 GPUs running on a single RDMA structure, with an estimated value of $3-4 billion. In terms of computing power, it's about 20 times more powerful than the setup used to train OpenAI's GPT-4.
Musk revealed that Grok 2 should be comparable to ChatGPT's latest model (GPT-4). Grok 3, currently in training, is expected to be released before December and could potentially be the world's most powerful AI at that time.
Discussing AI risks, Musk emphasized the importance of "pro-human" principles in AI development. He expressed concern about the goal of creating digital superintelligence far smarter than humans, which he believes is not beneficial.
Musk also shared his disagreement with former Google CEO Larry Page's views on AI and humanity's future. He stressed the importance of supporting humans over machines.
Regarding OpenAI, which Musk co-founded in 2015 but left in 2018 due to ideological differences, he criticized the organization for deviating from its original open-source, non-profit mission. Musk revealed he's considering legal action to understand how the organization he funded could pursue goals contrary to its initial purpose.