At a Twitter Space conference, Elon Musk held court on all things AI, dishing details on the newly announced xAI initiative (Elon Musk launched latest venture xAI, work closely with X Corp, Tesla) and Tesla’s own AI chip development. Though xAI and Tesla will collaborate, Musk emphasized they will maintain some distance.
Musk revealed that Tesla is currently developing its own chip, but it won’t be called a GPU or described as 100s or H100s. Instead, the company will name it something entirely original.
Furthermore, Musk shared that Tesla Dojo2, the company’s upcoming supercomputer, will focus on large models, marking a significant change in direction from previous versions. The function of Tesla HW 4 will be 3-5 times more powerful than its predecessor, HW 3, while HW 5 will be 4-5 times more potent than HW 4.
He also touched on Tesla’s supercomputer Dojo, which currently focuses on AI machine learning and computer vision training. However, Musk thinks they will center efforts on large models going forward.
“Dojo 2 will focus on large models,” remarked Musk.
Musk believes xAI and Tesla’s AI initiatives need to understand the physical world, not just the internet. He thinks Tesla’s driving data will aid progress in this endeavor.
When asked about collaborations with X.AI, Musk stated that the two companies would work together but keep some distance. He also revealed that AGI, or Artificial General Intelligence, will undoubtedly appear, and xAI will be involved in creating more competitive products than leading AI companies like Google, Microsoft, and OpenAI.
Musk raised concerns about unauthorized use of Twitter’s data for AI training and proposed that publicly available tweets could be used for training, like other companies. He also mentioned that Twitter’s dataset is suitable for “text training” and “image and video training,” and AI systems require more data than humans create.
Omar Qazi, also known as Whole Mars Catalog, asked Musk about using chips produced by NVIDIA or Tesla for data processing. Musk said that Tesla is developing its own chips and wouldn’t call them a “GPU,” but others could describe them as such.
Musk also discussed xAI’s development of AI language models that may not be politically correct, but which he believes can provide answers that are controversial but accurate. He believes that xAI needs to develop a technology that “understands the physical world, not just the internet,” and Tesla’s driving data will help it make progress in this area.
Musk also touched on Twitter’s vast data being used to train AI models without permission. He stated xAI will use publicly available Tweets, except private info, for training like everyone else. With AI’s voracious data needs, Musk thinks this is justified and apologized for recent usage restrictions affecting some developers.
If realized, Musk’s grand AI visions for xAI and Tesla could shake up the AI chip and model markets. But only time will tell if his ambitions manifest into reality.
Related Posts:
Tesla’s Feature end to end World Model: The Key to Full Self-Driving?
Tesla Dojo Supercomputer to Kick Off Production in 2023, Set to Radically Advance AI
Tesla AI Power Move at CVPR : Plans to Achieve 100 Exa-Flops by 2024