AI Competition Heats Up as xAI Closes Biggest Series B of All Time

0:00Today on the AI Daily Brief, XAI has raised$6 billion at a huge valuation. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes.

0:21Welcome back to the AI Daily Brief Headlines Edition, all the AI headlines you need in around five minutes. Today, because of some travel, I will not be doing the normal main episode, so this is actually going to be a quite extended headlines edition. there's been a lot to catch up on over the holiday weekend. We kick it off with a follow-up of a story that we were tracking when it was just rumor, which is of course Elon Musk's AI company behind the Grok chatbot, has officially completed this Series B round, and the details were largely what we knew. It was a$6 billion round at a post-money valuation of$24 billion.

0:53You might remember that at the very last minute, the round jumped in size from around$3 billion to$6 billion, given how much demand there was. If you're interested in the dynamics of that round, I did a whole episode about how if you are investing in the frontier model space, there are really such a small number of competitors that you can realistically make a bet on that you're sort of priced into paying whatever the market is going to pay. So who did participate in this? There are a lot of folks that you would expect. Andreessen Horowitz, Sequoia Capital. Fidelity was also an investor, as were the Saudis with Prince Alawid bin Talal and Kingdom Holdings.

1:24Of course, the big question is, did we get any interesting news in this announcement? Did they say anything about what's coming? The short answer is not really. They write,

2:02The funds from the round will be used to take XAI's first products to market, build advanced infrastructure, and accelerate the research and development of future technologies. So anyways, TLDR, no, we did not get anything new. However, that note about the funding being used to build out the infrastructure does have some interesting context. The information reports, Musk plans XAI supercomputer dubbed Gigafactory of Compute. Writes in Isikardizzi, Elon Musk has said publicly that his artificial intelligence startup XAI will need a whopping 100 ,000 specialized semiconductors to train and run the next version of its conversational AI Grok.

2:34To make the chatbot smarter, he's recently told investors XAI plans to string all these chips into a single massive computer, what he's calling a gigafactory of compute. So this was clearly a big part of this pitch. The information continues, in a May presentation to investors, Musk says he wants to get the supercomputer running by the fall of 2025. Musk claims that when completed, this would be at least four times the size of the biggest GPU clusters that exist today. It's clear that this infrastructure build-out is a key part of Elon's approach to catching up to rivals. At the same time, Microsoft and OpenAI have also discussed a$100 billion supercomputer that would be several times larger than this and contain millions of NVIDIA GPUs.

3:10Another interesting tidbit from the article, quote, XAI could partner with Oracle on the supercomputer. It has been talking to Oracle executives about potentially spending$10 billion to rent cloud servers over a period of years. XAI already rents servers with about 16 ,000 H100 chips from the company, making it Oracle's largest customer in that area. In terms of sooner updates, Musk has said that XAI is currently training Grok 2.0 on 20 ,000 GPUs, and this vision appears to push into multimodality. Importantly, all of this is separate from Dojo, which is a Tesla supercomputer, which is based on a chip that they built in-house and is focused on running the AI that works with self-driving capabilities.

3:45In addition to their in-house chips, Elon has also said that Tesla has access to 35 ,000 NVIDIA H100s to train itself driving AI, and plans to double that by the end of 2024. Amir Afradi at The Information writes about how this all fits into a larger trendline. He says, You're going to hear the term supercomputer a lot in the coming months. It's just a cool way of saying a large cluster of servers running NVIDIA chips, but without as many words. Anyone who's anyone in AI is going to be talking about their supercomputer, and we imagine that Elon Musk will be one of them. Amir continues, What strikes us is that his plan to get a cluster of 100 ,000 of NVIDIA's H100 chips by the end of next year will probably put him a full year behind OpenAI and Microsoft.

4:22You may or may not consider that to be a long time in AI land, but knowing how competitive Musk is, we won't be surprised if the Musk cluster ends up being even bigger than planned. Amir continues though, one wild card is NVIDIA CEO Jensen Huang and whether he wants to give Musk a head start or special allocation of NVIDIA's next flagship chip, Blackwell, at the expense of other firms. There is a precedent for this. NVIDIA a year ago aided smaller cloud providers with allocations of its H100 chips so they could better compete with bigger rivals like Amazon and Microsoft. In short, let the supercomputer race begin.

4:52So what was the chatter here? Well, a lot of it was just about the size. Andrew Gao points out that the$6 billion Series B is significantly larger than the previous biggest Series B, in fact doubling it. Others pointed out that Elon's key competitive advantage continues to be the relationship with Twitter. Robert Scoble writes, people tried to get me to leave Twitter after Elon bought it, saying he would screw up everything here. But I stayed because I knew Elon had a vision for the Everything app that would be compelling. Investors liked the pitch. Congrats, Elon. OpenAI just showed the world the Everything app.

5:21But I can't get it to tell me about the smoke in the sky. XAI has the necessary foundation to make a very compelling Everything app. That's already a high bar because OpenAI's is quite compelling already. Part of Elon's big promise is, of course, that Grok won't censor answers in the same way that the other big labs do. Speaking of the other big labs, a little bit of controversy as chief AI scientist at Meta, Yann LeCun couldn't resist taking to Twitter to rib Elon. In response to a post where Elon Musk wrote, join XAI if you believe in our mission of understanding the universe which requires maximally rigorous pursuit of truth without regard to popularity or political correctness.

5:54Yann LeCun responded, join XAI if you can stand a boss who claims that what you are working on will be solved next year, no pressure. Claims that what you are working on will kill everyone and must be stopped or paused, yay, vacation for six months. Claims to want a maximally rigorous pursuit of truth, but spews crazy ass conspiracy theories on his own social platform. When someone responded, not a good look, maybe explain what you have against Elon. Elon writes, I like his cars, his rockets, his solar panels, and his satellite network. I very much dislike his vengeful politics, his conspiracy theories, and his hype.

6:23Elon Musk eventually would respond, basically accusing him of being a proxy for Zuckerberg. Elon tweeted, Elon is quote, just following orders. Now, it may be just the fact that this poll was taken on Twitter, but when Bojan, formerly of NVIDIA, asked which AI lab people would most like to work at, XAI absolutely crushed. With more than 91 ,000 votes, XAI was at 51.4 % compared to OpenAI's 31.2%, compared to Google DeepMind's 10.8%, compared to Meta AI's 6.6 % down near the bottom. Elon responded, but he added a snapshot before he did, and when there were about 4 ,500 votes before Elon got there, the results were XAI at 40%, OpenAI at 29%, Meta at 17%, and Google DeepMind at 14.

7:05So XAI was still leading. Again, we are on Twitter slash X, so there's going to be some bias here, but still it's pretty dramatic relative to what I might have expected. Hello friends, before we get back to the episode, I want to tell you about something special I'm doing on Superintelligent this June. Super is of course our platform for AI learning, and I've heard from a lot of you that you really want something for a true AI beginner, someone who's really just getting their feet wet with these tools. So what I'm going to do is put together basically a course that sits on top of and uses super intelligent tutorials and lessons, but where I hand guide you through around 10 different lessons and how-tos that I think once you complete them will have you ahead of 80 % of the other people who are just starting to use AI right now.

7:44If you are interested in this learning experience, go to besuper.ai and sign up using code JUNE. You'll get 25 % off your first month and I'll automatically add you to that AI for beginners group. That's bsuper.ai, discount code June. See you there. Speaking of OpenAI, in the wake of all of its super alignment researchers leaving and that team being disbanded, the company has announced that they've created a new committee to oversee safety and security. OpenAI says the committee would comprise several of its board members and employees, including CEO Sam Altman. It would evaluate and further develop the company's processes and safeguards over the next 90 days.

8:19Now, the part of the post that the New York Times picked up was captured in the headline, OpenAI says it has begun training a new flagship AI model. Cade Metz writes, OpenAI said on Tuesday that it has begun training a new flagship artificial intelligence model that would succeed GPT-4. The startup said in a blog post that it expects the model to bring the next level of capabilities as it strives to build AGI. Regarding the new safety and security committee, OpenAI wrote, while we are proud to build and release models that are industry-leading on both capabilities and safety, we welcome a robust debate at this important moment.

8:47A couple things that I think are interesting about this. First of all, there's no universe in which OpenAI hasn't been training post-GPT4 models for a very long time. I think the presumption, as the New York Times is making it, that this is just starting now and that it could be 9 to 12 months before they come out is totally absurd and a misreading of the whole situation. There's simply no way that this hasn't been ongoing. I mean, I remember a year ago when I was discussing this and asked people what they thought, and there was a ton of conviction that training was already happening just without discussion.

9:15The second part, though, which is even more interesting, has to do with a conversation that we had starting over last weekend, prompted by Tyler Cowen's op-ed in Bloomberg that the AI safety movement had died. Now his point, hold aside the hyperbolic headline, was that there had been a major shift, and for whatever reason, people just weren't paying as much attention to AI safety as they had been, call it, six months ago. It feels to me a little bit like OpenAI is testing the waters around really what public sentiment here is. Part of the reason that they might be giving themselves these 90 days to figure out what the right approach to AI safety is, is to actually gauge public sentiment during that time.

9:47Basically, are they going to design this entirely based on what they actually think, versus what do they have to do to appease external pressure? Certainly those concerned with AI safety haven't just crawled into some hole. The Economist, for example, ran an op-ed by Helen Toner and Tasha McCauley, who are two former OpenAI board members, with Helen especially being at the center of the whole Sam-firing-rehiring saga. And the op-ed is called, AI Firms Mustn't Govern Themselves. Can private companies pushing forward the frontier of a revolutionary new technology be expected to operate in the interest of both their shareholders in the wider world.

10:18When we were recruited to the board of OpenAI, we were cautiously optimistic that the company's innovative approach to self-governance could offer a blueprint for responsible AI development. But based on our experience, we believe that self-governance cannot reliably withstand the pressure of profit incentives. With AI's enormous potential for both positive and negative impact, it's not sufficient to assume that such incentives will always be aligned with the public good. For the rise of AI to benefit everyone, governments must begin building effective regulatory frameworks now. In my opinion, we have about a seven or eight month period now where these debates are going to be had without any real stakes.

10:48What I mean by that is that I don't think that the stakes are going to present themselves until after the U.S. elections, and specifically until the new Congress is sworn in. I think the reality is that as much as other nations hold safety summits and things like that, it will be the U.S. regulatory regime that will have the biggest impact in how these questions play out, in what the relationship between the private sector and industry actually is, and I just don't see that happening before the elections are done. In the meantime, the competition between the big tech companies continues. As we get closer and closer to Apple's WWDC, the rumor mill is in full swing, with one of the most listened-to sources being Bloomberg's Mark Gurman.

11:23His latest on this is called Apple Bets That Its Giant User Base Will Help It Win in AI. And the TLDR on this is Don't Expect Some Big Crazy Impressive Thing. The header sentence of this piece, though Apple's first set of modern AI features won't be as impressive as rival offerings, the company is betting that its massive customer base can give it an edge. At Apple's developer conference next month, the company will unveil a different approach to artificial intelligence, focusing on tools that ordinary consumers can use in their daily lives. The idea is to appeal to a user's practical side and leave some of the more whiz-bang features to the other companies.

11:55Apple, Gurman points out, is in a challenging position. It needs to convince consumers and investors that it's doing exciting things in AI. But the company is following major AI announcements from Microsoft, Google, and OpenAI, which have stolen the spotlight. According to Gurman, AI will be the key focus of the WWDC, but it seems like it's going to be a lot AI integration into existing core apps. Gurman writes, There are several new capabilities in the works for this year, including ones that transcribe voice memos, retouch photos with AI, and make searches faster and more reliable in the Spotlight feature.

12:22The Siri personal assistant will get an upgrade as well, with more natural-sounding interactions based on Apple's own LLMs, and there's also a more advanced Siri coming to the Apple Watch. One standout feature, writes Gurman, will bring generative AI to emojis. The company is developing software that can create custom emojis on the fly based on what users are texting. That means you'll suddenly have an all-new emoji for any occasion. Gurman points out, however, that this is all purely catch-up, his words. There's no leapfrogging here. There's also no Apple-designed chatbot. Gurman also says that a partnership between OpenAI and Apple will be a component of the WWDC announcement.

12:54This all strikes me as really difficult, frankly. On the one hand, it is absolutely the case that most people are going to have their first interactions with AI in the context of applications they already know and use on devices that they already have. And in that way, I wouldn't count Apple out. At the same time, the seismic gap between what it appears they'll announce and what other companies are already doing seems very likely to me to make Apple feel extremely far behind when it comes to consumer popular sentiment. And that's to say nothing of Wall Street, which I can't imagine being particularly excited about this.

13:27If I'm right, the question will become whether the fact that Apple has all those installs already, and people they can push AI to, will beat out the fact that they're still running behind? Or, on the other hand, do the folks who are early adopters just need something more exciting? Luckily, we don't have to wait very much longer for that, and so we'll finally be able to talk about what is actually here rather than just future rumors. A couple more before we get out of here. Grok, not Elon's Grok, but the G-R-O-Q Grok, is apparently out raising$300 million. You might remember Grok is the extremely fast inference chip that has really impressed people with their demos which you can find online.

14:03The information writes, Started nearly eight years ago by Jonathan Ross, one of the inventors of Google's specialized chip called the Tensor Processing Unit, Grok has positioned itself as a cheaper and faster alternative to NVIDIA chips. Grok designs chips for AI inference, as opposed to chips that help companies like OpenAI train new models. The information writes, Even with more money, Grok will face an uphill battle as it aims to get developers to switch off NVIDIA's chips and the accompanying software they're familiar with, known as CUDA. Much larger companies, including Amazon, Google, and Microsoft, have been trying for several years to develop an AI server chip to wean their cloud customers off NVIDIAs, but they've made limited progress.

14:36Over in Google land, the company is once again scrambling to manually remove all those weird answers that were coming up in search, things like putting glue on pizza and suggesting that users eat rocks. And as The Verge writes, it's an odd situation since Google has been testing AI overviews for a year now. And CEO Sundar Pichai has said the company served over a billion queries in that time. But Pichai has also said that Google has brought the cost of delivering AI answers down by 80 % over that same time, driven by hardware, engineering, and technical breakthroughs. It appears that kind of optimization might have happened too early before the tech was ready.

15:06Said one AI founder who wished to remain anonymous, a company once known for being at the cutting edge and shipping high-quality stuff is now known for low-quality output that's getting memed. So as always, tons going on in the world of AI. I'm sure we will have more to discuss later this week, but for now, that is going to do it for the AI Daily Brief. Appreciate you listening or watching as always, and until next time, peace. Thank you.