Elon Starts another Company - AI this time

[Yazata]

He's been talking about it for some time. He was one of the founders of OpenAI, but later left. But Elon's still vitally interested in the subject, as illustrated by his self-driving cars and his Teslabots.

So here's Xai

They say, "The Goal of Xai is to understand the true nature of the universe" (nothing like starting out small with a modest easy-to-achieve goal).

But it does fit in with Elon's vision, which seems to be to make science fiction real.

https://x.ai/

[Magical Realist]

Well, who better to steer the course of AI development than someone who appreciates the possible dangers of a future AI takeover:

"Elon Musk has issued another warning about the risks posed by advanced artificial intelligence.

Speaking at the WSJ CEO Council Summit, Musk said AI had the potential to assume control of humanity. The billionaire told the Journal's Thorold Barker: "It's a small likelihood of annihilating humanity, but it's not zero."

He said there was also a chance AI would assume control for "the safety of all the humans" and effectively become some sort of "uber-nanny."

"There is a risk that advanced AI either eliminates or constrains humanity's growth," he said. Adding that super-intelligence was a "double-edged sword."

"If you have a genie that can grant you anything, that presents a danger," he said.

Musk has been wary of the powerful technology for some time despite investing in AI development within his own companies. In March, he signed an open letter calling for a six-month pause on advanced development. The letter cited potential risks to humanity and society, which included the spread of misinformation and widespread automation of jobs."---
https://www.businessinsider.com/elon-mus...20zero.%22

[Yazata]

Well, who better to steer the course of AI development than someone who appreciates the possible dangers of a future AI takeover:

"Elon Musk has issued another warning about the risks posed by advanced artificial intelligence.

Speaking at the WSJ CEO Council Summit, Musk said AI had the potential to assume control of humanity. The billionaire told the Journal's Thorold Barker: "It's a small likelihood of annihilating humanity, but it's not zero."

He said there was also a chance AI would assume control for "the safety of all the humans" and effectively become some sort of "uber-nanny."

"There is a risk that advanced AI either eliminates or constrains humanity's growth," he said. Adding that super-intelligence was a "double-edged sword."

"If you have a genie that can grant you anything, that presents a danger," he said.

Apparently that's Elon's reason for giving AI the task of understanding the true nature of the universe.

Elon did a 'Twitter Space' about Xai last night (I didn't tune in) and Reuters describes his thinking this way:

Rather than explicitly programming morality into its AI, Xai will seek to create a "maximally curious" AI. "If it tried to understand the true nature of the universe, that's actually the best thing that I can come up with from an AI safety standpoint," Musk said. "I think it is going to be pro-humanity from the standpoint that humanity is just much more interesting than not-humanity."

https://www.reuters.com/technology/elon-...023-07-12/

"Life, the Universe, and Everything!"

https://twitter.com/elonmusk/status/1679546977565184002

[Yazata]

X.ai was registered in March 2023 and announced in July. It apparently already has $135 million in capital from Elon and several other investers. And it's just announced a $1 billion funding round. So it might become a "unicorn" in less than a year. (A privately held startup with valuation above a billion.)

When asked whether X.ai will compete with Tesla's huge AI effort, Elon said this:

"There were just some of the world’s best AI engineers and scientists that were willing to join a startup but they were not willing to join a large, sort of relatively established company like Tesla, so I was like, OK well, better it’s a startup that I run than they go work somewhere else. That’s kind of the genesis of X.AI.”

CNBC says:

"People working on X.AI include alumni of DeepMind, OpenAI, Google Research, Microsoft Research, Twitter and Tesla. They’ve worked on projects including DeepMind’s AlphaCode and OpenAI’s GPT-3.5 and GPT-4 chatbots, according to LinkedIn profiles."

Again CNBC:

"Earlier this year, Musk reportedly secured thousands of high-powered graphics processing units (GPUs) from Nvidia, the kind of chips necessary to build a large language model."

I'd heard that, but the story I heard (perhaps wrong) was that it was Tesla's AI effort that secured them. It wasn't just thousands, it was many thousands.

These super-high-powered graphics chips, originally created to power video games, have proven to be vital to powering AI as well. There are only a few manufacturers worldwide, with Nvidia the leading supplier. Hence the explosive rise in Nvidia's stock price. Obtaining the necessary GPU chips has been a major limiting factor on new AI efforts.

https://www.cnbc.com/2023/12/05/elon-mus...lion-.html

[confused2]

It seems (an AI told me) GPU chips are only used in the learning phase so lots of GPUs means (I guess) lots of learning.

GPUs are often used during the training phase of large language models like myself. However, once the model has been trained, it's essentially a static set of data and algorithms that doesn't require the use of GPUs for inference (or the process of using the model to generate responses). Instead, we rely on high-performance CPUs and other server-grade hardware to generate responses to user queries as quickly as possible. So, to answer your question, I'm not currently running on GPUs!

Edit.. As I understand it the GPU chips are good for handling the interconnectness of everything in the training phase - so the AI can relate invisible [Chinese fighting] fish to 'something'. If Elon wants to add new stuff in real time he might still need lots of GPUs at run time.

One potential challenge with real-time learning is the amount of computational power it requires - GPUs are great for processing large amounts of data quickly, but even they would struggle to handle the sheer volume of data that a real-time learning AI system might require. Still, it's an exciting idea with a lot of potential!

[Yazata]

More on how X.ai sourced those thousands of GPU chips -

Larry Ellison (Oracle Chairman and friend of Elon) said on an Oracle call with investors Monday,

“We got enough Nvidia GPUs for Elon Musk’s company, xAI, to bring up the first version, the first available version of their large language model called Grok... They got that up and running. But boy, did they want a lot more. Boy, did they want a lot more GPUs than we gave them.”

So apparently they didn't come from Tesla's big Nvidia GPU haul, which I believe are mostly being used to train the Tesla full-self-driving neural networks. X.ai's GPU's seem to have come from Oracle.

[Yazata]

Elon wrote today on X about why he favors what he takes to be the epistemology of physics and why he's applying it to Xai's AI:

Physics has many powerful tools of reasoning to understand and predict reality. Those tools, like first principles analysis and thinking in the limit, are broadly applicable to anything, in my experience.

With Grok, @xAI is attempting to create an AI that reasons from first principles, which is fundamental if you care about getting as close to the truth as possible.

The acid test would be reaching a conclusion that is correct even if it is extremely unpopular, which means being right even when the training data is almost entirely wrong.

For example, Galileo concluded, after observing the moons of Jupiter from a telescope he engineered, that it was far more probable that Earth revolved around the Sun than the other way around. This view was so unpopular that he was forced to recant and placed under house arrest!

If you had trained an LLM on material back then, it would’ve given you the popular, but wrong, explanation. Due to social and legal pressure, it likely wouldn’t even acknowledge the possibility that the Earth revolved around the Sun.

For AI to help us understand the true nature of the universe, it must be able to discard the popular, but wrong, in favor of the unpopular, but right.

[Yazata]

X.ai was registered in March 2023 and announced in July. It apparently already has $135 million in capital from Elon and several other investers. And it's just announced a $1 billion funding round. So it might become a "unicorn" in less than a year. (A privately held startup with valuation above a billion.)

xAI recently completed another funding round that appears to value the company at an impressive $18 billion!

https://www.msn.com/en-us/money/companie...r-BB1m5AoR

"Previous reports from Bloomberg and other outlets including TechCrunch and The Information suggest Musk’s AI startup is raising $6 billion at a valuation of around $18 billion, up from reports of a deal worth around $3 billion at a $15 billion pre-money valuation due to intense investor interest.

Participants in the round reportedly include venture capital giant Sequoia Capital, a Silicon Valley titan with a long history of backing Musk’s projects, including the PayPal precursor X.com, tunneling business The Boring Company, rocket venture SpaceX, as well as supporting the billionaire’s efforts to take over Twitter, now X, in 2022."

And in other xAI news, xAI is reportedly in talks with Oracle to finalize a huge $10 billion dollar deal for xAI to use Oracle cloud servers.

"According to Reuters, xAI is in talks with Oracle executives about a $10 billion deal to rent cloud servers from the company. The proposed deal would be similar in size to the deals OpenAI and Anthropic struck with Oracle’s bigger cloud competitors. This would make xAI one of Oracle's largest customers. This deal for xAI will accelerate its ability to rent chips and help it reach 100,000 GPUs by the second half of 2025."

[Yazata]

Elon says:

"Nice work by @xAI team, @X team, @Nvidia & supporting companies getting Memphis Supercluster training started at ~4:20am local time.

With 100k liquid-cooled H100s on a single RDMA fabric, it’s the most powerful AI training cluster in the world!"

https://x.com/elonmusk/status/1815325410667749760

[Secular Sanity]

Just finished watching his interview with Jordan Peterson. It was pretty good. 

https://x.com/jordanbpeterson/status/181...8703090085