Elon Musk Announces xAI: Here's Everything We Know So Far

0:00Today on the AI Breakdown, we're looking at everything we know about Elon's new project x.ai. Before that on the brief, Adobe testifies before the Senate, Google Bard releases a set of updates, and much, much more. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube channels, Discord, and newsletter. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five-ish minutes or less. The last 24 hours have been huge for AI news. Obviously, Elon's XAI announcement is a big one, but that's what the main AI breakdown is going to be focused on today, so come back soon for that.

0:38Where we're going to start is actually with Google. The company announced a number of big updates for its BARD LLM, and I think in many ways, the way that I would summarize the connective tissue of all of these updates is that they're trying to add utility, functionality to the underlying powers of this tool. And you'll see as I dig into the updates specifically what I mean. First up, BARD is now available in more languages. BARD can be used in Europe now, and it can be used in over 40 languages that include Arabic, Chinese, German, Hindi, and Spanish. Brazil also comes online as a place where BARD is available.

1:13Next up, they've announced a feature where instead of just reading what BARD says about a particular topic or question, you can actually listen to it. They point out that this could be helpful if one is trying to hear the correct pronunciation of a term, if the response to the query is a type of language that is meant to be heard, such as a poem or a script, or frankly, if people just want to listen on the go rather than read. They've also announced more granular controls of BARD's responses. There's now one-click updates along three vectors. The first is simply complexity. When BARD produces a response, you can go down, click the Modify Response button, and select Simpler.

1:48Another vector for transformation under this Modify menu is making the response longer or shorter. Finally, a third way to modify the response is the tone, with Bard giving you the ability to, with a single click, make the response either more professional or more casual. From a functionality standpoint, they've introduced a feature by which people can pin and rename conversations. This is obviously valuable if there's a topic that you need to return to later after the session that you're currently in. And one that they were really excited to show off was new ways to export code to other places where developers live.

2:20Specifically today, they announced a new feature that allows developers to export Python code to Replicit rather than just locking them into Google Collab. It seems likely that that sort of export code function is just going to expand from here. BARD now allows users to share the responses that they like with their friends or family or colleagues. And one of the big ones in terms of really expanding the functionality is a first step into multimodality. They've brought the capabilities of Google Lens into BARD so now you can input a picture as part of the prompt. This goes way beyond the sort of Google image search that you might be used to, and instead allows you to, for example, post an image and ask for a caption.

2:56Or you could post a beautiful scene and ask for a poem about that scene. They don't use the word multimodal in their announcement, but that's effectively what this is, a first step in that direction. In many ways, this is the biggest substantive change, and the thing that makes this iteration of Bard most different from its competitors. Still, overall, this combination of features definitely increases the uses and functionality of Bard in ways that I think will likely lead to much more usage going forward. Next up, some news from the intersection of AI policy and AI companies. The Senate Judiciary Committee Subcommittee on Intellectual Property held a hearing yesterday called Artificial Intelligence and Intellectual Property.

3:32The focus was on copyright, and they had a number of expert witnesses, including the General Counsel at Universal Music Group, a professor of law at Emory University, the EVP General Counsel and Chief Trust Officer at Adobe, the Head of Public Policy for Stability AI, and a concept illustrator and fine artist from San Francisco. One of the interesting things to come out of that was testimony from Adobe that they would support giving artists the ability to opt out of AI training on the things that they've created. Dana Rao said, We believe creators should be able to attach a do not train tag to their work.

4:02With industry and government support, we can ensure AI data callers read and respect this tag, giving creators the option to keep their data out of AI training data sets. Rao also went even farther saying, we believe artists should be protected against this type of economic harm, and we propose Congress establish a new federal anti-impersonation right that would give artists the right to enforce against someone intentionally attempting to impersonate their style or likeness. This seems extremely challenging to me. Something like the do not train tag I think makes a little bit more sense. But a law that tries to define if and how AI-generated art is or isn't too close to another artist's style is an incredibly difficult, incredibly subjective thing to actually implement.

4:41It sort of assumes that every artist is a complete individual unto themselves, rather than an interesting combination and remix of all the influences they have. Obviously, it would come down to the details, but it's hard not to imagine how a law like this wouldn't just end up creating a massive honeypot for litigators trying to argue that their clients were being ripped off by AI, whether or not it was exactly true. But I still do think it's interesting and reflective of where the conversation is right now. Staying on the theme of AI policy for a moment, Vice President Kamala Harris yesterday held a conversation with civil rights leaders and consumer protection advocates around artificial intelligence.

5:15The discussion was supposed to be a working session designed to cover and enumerate AI risks to vulnerable populations, which could include people like seniors. And although people were pleased to see this level of attention being given to the issue, the specific way in which Vice President Harris described AI didn't exactly inspire confidence. And I think the first part of this issue that should be articulated is AI is kind of a fancy thing. First of all, it's two letters. It means artificial intelligence. We also got announcements yesterday about new commercial AI products that are available.

5:48Shopify is launching an AI assistant for merchants as a for example. The assistant, which they call Sidekick, will help merchants with questions around sales trends, product details, and more. Over in the world of Meta, the Financial Times is reporting that they are moving forward with plans to launch a commercial version of their Llama model. and NVIDIA has announced a$50 million investment into biotech company Recursion around AI drug discovery. Now to get a sense of just how hot to trot the market is for AI right now, following the announcement, Recursion stock was up 80%. Shares of NVIDIA, meanwhile, went up more than 2%.

6:22All right, guys, that is going to do it for today's AI Breakdown Brief. If you're enjoying it, please subscribe to the channel, or if you're listening to this, go check out the YouTube channel. You can get a link at breakdown.network, and I'll be back soon with the main AI breakdown. Elon Musk is getting into the AI game with a new company he's calling X.AI. What's the deal? What's it doing? And can he compete with giants in the space already like OpenAI? Welcome back to the AI breakdown. Elon Musk once tweeted the cryptic message. In retrospect, it was inevitable. It's hard not to feel similarly about his entrance into the wild, wacky world of AI.

7:03Now, to be fair, Elon is not really some new entrant into the AI space. As we'll get into in this, he has been around for a long time. In fact, he was one of the original funders of OpenAI, a point of some contention given how that company has evolved. What happened yesterday, however, is that Elon Musk announced a new company, XAI, which is meant to be his big, full push into the space. The details so far are incredibly scant. The website, which is at x.ai, reads, The goal of XAI is to understand the true nature of the universe. You can meet the team and ask us questions during a Twitter Spaces chat on Friday, July 14th.

7:40Our team is led by Elon Musk, CEO of Tesla and SpaceX. We have previously worked at DeepMind, OpenAI, Google Research, Microsoft Research, Tesla, and the University of Toronto. Collectively, we contributed some of the most widely used methods in the field, in particular the atom optimizer, batch normalization, layer normalization, and the discovery of adversarial examples. We further introduced innovative techniques and analyses such as Transformer XL, autoformalization, the memorizing transformer, batch size scaling, and U-Transformer. We have worked on and led the developments of some of the largest breakthroughs in the field, including AlphaStar, AlphaCode, Inception, Minerva, GPT 3.5, and GPT-4.

8:17In short, what this web page says is we're not going to tell you what we're doing. but you damn well better believe that we're going to do it well. Now, before we get into XAI and what people's reaction to it is, let's zoom back just a little bit to talk about Elon's contentious relationship with OpenAI. Elon has been, let's say, pretty outspoken since the launch of ChatGPT launched OpenAI into mainstream recognition. In May, Elon did an interview with CNBC's David Faber, and in it, he had some pretty harsh words for OpenAI. One of his claims was that he was the, quote, reason that OpenAI exists, citing his early investment in the company, which was at that time a non-profit.

8:54He said that he was instrumental in recruiting scientists at the beginning of that company's life. And he even said, quote, I came up with the name. Now, initially, Elon had committed something like a billion dollars to the endeavor, although ultimately he ended up investing more like 50 million. As OpenAI has changed from a non-profit to a model where there is a capped profit and the for-profit entity underneath is owned by the non-profit, he's been a little bit skeptical. He's repeatedly asserted that Microsoft has control over OpenAI, and in general thinks the model is dubious. He's also numerous times suggested that OpenAI does not put enough emphasis on AI safety.

9:27Elon did sign that letter from March asking for a six-month pause, even though he said that he knew that one, it would never happen, and two, even if it did, it wouldn't really be effective at accomplishing its goals. Now, the members of the team who launched x.ai were very mum on Twitter as well. They really hammered this idea of exploration. Greg Yang writes, Finally launched X.AI. The mathematics of deep learning is profound, beautiful, and unreasonably effective. Developing the theory for everything for large neural networks will be central to taking AI to the next level. Conversely, this AI will enable everyone to understand our mathematical universe in ways unimaginable before.

10:03Math for AI and AI for math. The main XAI account also hammered this theme of exploration and finding answers. They tweeted, What are the most fundamental unanswered questions? To that, Elon responded, And what are the most fundamental unknown questions? Once you know the right question to ask, the answer is often the easy part, as my hero Douglas Adams would say. Now when it came to the discussion, a lot of it focused around the team. The negative side of that, as pointed out by Rachel Metz from Bloomberg, was that all of the team members listed on the website were men. And people wondered if the goal of this organization was to really explore the universe, might they need a team that represented a broader cross-section of the universe?

10:42The flip side, however, was how much the team communicated the seriousness of the project. Jim Fan from NVIDIA wrote, It's got an all-star founding team. I'm really impressed by the talent density. Read too many papers from them to count. He goes through a number of the team members and talks about their contributions, such as Igor Babushkin's AlphaStar, which was DeepMind's StarCraft player that beats Human Champions, Greg Yang's Deep Learning Theory, and many more. Now, one of the big topics of conversation was how this move related to all the other moves that Elon's been making recently. Alex Felaitis pulled it together this way.

11:16He wrote, Elon's master plan is becoming clear. Today, Elon announced the formation of a new AI company called XAI. It's an important step towards his endgame of creating X, the everything app. Here are the tools that Elon has at his disposal. 400 million Twitter accounts with years of text to train his AI agents on. Starlink satellite network, which can bring internet to all corners of the planet. XAI, which boasts a team of the top AI researchers on the planet. If you've been looking at Twitter through the lens of just being a social network, then you're missing the plot. Similar to how Elon took a set of deliberate steps in launching Tesla, we're seeing a similar master plan play out with X.

11:51Start viewing things through this wider lens, and his recent actions don't seem crazy. They seem deliberate. We need to think from first principles. What will define the future of the internet? The companies that will have the most control will be the ones who possess the following. The most data, Twitter and Tesla. the most powerful infrastructure, Tesla Global Charging Network plus factories plus Starlink Network, the best AI talent, XAI and Neuralink, and the most capital, richest man in the world. Elon is assembling all of the pieces. Now he just needs to put them together. Without necessarily going as deep on this whole theory, Jim Phan did say something similar.

12:24He wrote, XAI led by Elon Musk is the latest heavyweight player in AI. I see a few unique strengths in Elon's ecosystem. Lots of multimodal data on Twitter, Dialogue text, images, and a growing collection of long videos. XAI is the only AI company that has direct and legal access to such an enormous and daily expanding corpus. Tesla FSD team has years of experience in building huge training clusters such as Dojo, not to mention the tons of high-quality Tesla fleet data for machine perception. TeslaBot would be the physical embodiment of the XAI brain if they ever connect the two. Now, when Elon bought Twitter, for what many considered a pretty inflated price, kind of the only theory for why that made sense to many people was that he bought it for that data to train artificial intelligence.

13:08As one Quora user wrote, According to sources close to Musk, the billionaire is fascinated by the intricacies of human behavior and is keen to understand how people interact with one another on social media platforms. Musk believes that by analyzing the massive amounts of data generated by Twitter users, AI systems could be trained to predict human behavior more accurately than ever before. And I think this is one thing that's worth noting. When you get the trove of Twitter data, you're not just getting information raw. You're also getting a set of interactions. Twitter is based on conversations and human interfacing.

13:39That's a very different type of data set than just a bunch of research papers as a for example. The XAI announcement and broader theory of how it comes together with the other pieces of the kingdom also help explain potentially why Elon has seemed so aggressive as it relates to other companies training their models on Twitter's data. Remember in April, after Microsoft dropped Twitter from its advertising platform because they refused to pay Twitter API fees, Elon tweeted they trained illegally using Twitter data. Lawsuit time. Now, when it comes to other commentary in the past six months or so about how Elon thinks about the AI space, really two areas that he's brought up frequently time and time again as areas of concern.

14:19The first is a fear of something that he referred to as woke AI. Time magazine wrote an article, Elon Musk is bringing the culture wars to AI. The article referenced a tweet from December where Elon said the danger of training AI to be woke, in other words, lie, is deadly. He also did an interview with Tucker Carlson in April where he said that he was working on something that he called TruthGPT. In that interview, we got a lot of what Elon has said publicly about his thoughts on how AI should be designed. And in some ways, he tried to connect the dots between that anti-woke AI, TruthGPT, with his view of AI safety.

14:53He said in that interview, I'm going to start something which I call TruthGPT, or a maximum truth-seeking AI, that tries to understand the nature of the universe. I think this might be the best path to safety in the sense that an AI that cares about understanding the universe is unlikely to annihilate humans because we are an interesting part of the universe. Elon built on these themes in the Twitter spaces he held yesterday with Congressman Ro Khanna and Congressman Mike Gallagher, in which he said, If I could press pause on AI or advance superintelligence systems, I would. However, Elon again said that that was unrealistic, and so instead he wanted to try to build a different type of AI.

15:27He repeated again the idea that the AI that he wanted to build, or grow as he put it, was one which would be maximally curious and maximally truth-seeking. He said, quote,

15:47He then went on to say that he thought that this was a better approach than trying to explicitly program morality into AI. The reason being the morality inversion problem, or as he put it, the Waluigi problem, where if you program morality, once you've even addressed the question of which morality and whose morality, programming one type of morality means you also get the inverse of it. In other words, if you program Luigi, you get Waluigi as well. Now theoretically, we'll get more information on Friday in that spaces that they advertised about exactly what XAI is doing. However, he definitely made it seem from this conversation at least that a big part of the focus was this question of AI safety.

16:23and why he didn't think the approaches currently happening were likely to work. The other interesting part of that discussion with Elon was the two congressmen asking for his take on China's position vis-a-vis AI, given that he has had more communication recently with CCP leadership than basically anyone else in America. Elon said that he thought that China understood the risk that a superintelligent AI would undermine CCP control, and that he believed that they would be interested in being a part of a transnational global regulatory regime as related to AI. Now, the congressman, I must say, expressed some significant skepticism of this.

16:56In fact, Congressman Mike Gallagher said, I'm just not sure they aspire to be on team humanity. I see them acting as if they're on team genocidal communism mostly. Now, holding those comments aside, earlier today, China did unveil its latest rules around generative AI, which will include a licensing regime that requires companies that are releasing these models to get a license to do so. As TechCrunch puts it, first and foremost, the rules require generative AI providers to adhere to core socialist values, which prohibit everything from pornography and terrorism to racism and content that threatens China's national security.

17:27Anyway, ultimately, until we get more information, it's just a big realm of speculation. We've got all the theories that you heard before. We've got people like Chroma founder Anton saying XAI might be trying to solve math. The one thing that I think people agree on is that this seems like a big effort. Notion co-founder Chris Prucha writes, pretty big deal. This might be a true counterweight to open AI. So at this point, I will ask, what do you guys think? What do you think Elon and XAI's unique contribution to this space will be? Will it be a totally different model of training, a different approach to AI safety, part of some broader master plan, or just Elon once again falling into main character syndrome?

18:08Use the comments to discuss it. Come join us on the AI Breakdown Discord, which you can get to at bit.ly slash AI Breakdown. I can't wait to hear what you guys think. For now, XAI is here and we're going to learn a lot more about it in the days and weeks to come. So until next time, guys, peace.