| Can You Grok It? | By Jeff Brown, Editor, The Bleeding Edge | Jeff’s Note: We’re coming up fast on my Overnight AI Gains Summit with “Market Wizard” Larry Benedict… | Have you ever wanted to double your money in less than a day? | If you’re skeptical, that’s natural. But with “zero-day” trades, regular people are doing just that… and more. | “Zero-day” trades last less than 24 hours… which means they can hand you some of the biggest, fastest gains of your life. And when you combine them with big trends like AI… hold on tight. | That’s why, tomorrow night at 8 p.m. ET, I’ll sit down with Larry to share exactly how you can add these trades to your portfolio. | Larry is one of the few “Market Wizards” in the world… and his hedge fund was ranked in the top 1% by Barron’s on several occasions. That’s why I knew exactly who I wanted to talk to when I heard about these trades. | So don’t wait… Go right here to reserve your spot at this event with one click. We hope to see you tomorrow night… | | It’s not artificial general intelligence (AGI). | Not yet, at least. | Last night at 8 p.m. ET, Elon Musk and his team at xAI held a launch event for Grok 3, its most advanced reasoning AI model. For anyone curious to view it and has an hour to spend, you can find it here. | I can save everyone some time, though… | xAI’s release of Grok 3 is nothing short of incredible. | Alphabet (GOOGL), Meta (META), Microsoft (MSFT), OpenAI, and Anthropic must all be shaking in their boots right about now. | Leading in All Categories | Grok is a rapidly evolving AI model that has already learned to identify its own mistakes and self-correct. | And it has the ability to “think” and consider all possible solutions, iterate, and optimize for the best one. | Musk even commented that “ We’re seeing the beginnings of creativity,” something that artificial intelligence (AI) critics have always said is impossible. | The significance of what xAI just released is hard to overstate. In one fell swoop, xAI and its Grok 3 AI model went from something that many in the industry wouldn’t even acknowledge as a player just six months ago, to state-of-the-art in all categories. | Grok 3 is now leading the race to AGI …. | | Source: lmarena.ai | It’s not that xAI is on top that has its competitors nervous… It’s how fast xAI got there. | OpenAI has been at it in earnest since 2019 when it had its first formal funding round, raising $1 billion. | Comparatively, xAI started developing in just 2023 and had its first institutional round in 2024. Just look at the chart below. | | Source: xAI | Look at the remarkable trajectory of how quickly xAI developed a state-of-the-art frontier model. This is why the other major players developing frontier AI models are so jittery. | What’s xAI’s secret? | A Colossal Feat | Aside from Musk’s ability to recruit and lead the very best engineers who are mission driven, the secret is Colossus. | | More specifically, it was the remarkable feat of engineering, troubleshooting, and innovation to spin up the world’s most powerful GPU supercluster in just months. | Shown in the image above is Colossus, a hyperscale “AI factory” running 200,000 NVIDIA GPUs. xAI stunned the industry last year when it spun up 100,000 GPUs in just 122 days, a feat that was previously unthinkable. | And last night, the xAI team shared that they doubled the compute capacity in just 92 days, totaling 200,000 GPUs. | That’s start to finish in just seven months – unbelievable. | How was it even possible? | xAI thought outside the box. It looked for a large, abandoned factory that could be repurposed as an AI data center. It found one in an Electrolux factory that was no longer in use sitting outside of Memphis. | But the building had only 15 megawatts of power. xAI needed 120 megawatts. So xAI purchased its own natural gas-fueled generators and stacked them on one side of the building. What the power company couldn’t provide, xAI simply produced itself. | On the other side of the building are cooling systems to keep the building temperatures down for performance and to avoid overheating. Even the GPUs are all liquid cooled. | One of the biggest challenges was dealing with the power fluctuations of a GPU supercluster like Colossus. In just 100 milliseconds, the power demands can either spike or fall, which is not something a power company can manage… | So xAI incorporated Tesla megapacks – battery-based power storage systems (the four rows of white objects in the Colossus picture). The megapacks acted as a buffer to smooth out the erratic power demands. | The result? A frontier AI model that beats anything released so far by OpenAI, Meta, Alphabet, Anthropic, DeepSeek, or anyone else. | | Shown by the blue bars above are the Grok 3 and Grok 3 mini-performance compared to the closest competitors. | But where the results get really interesting are with the Grok 3 models that incorporate “reasoning” and test-time compute… | | Time to Think | An easy way to think of test-time compute is like giving the AI model more time to think. | More time to think is simply more computational time to consider all possibilities and come up with the best answer. | Reasoning models are currently the most advanced frontier models. And Grok 3 is now outperforming all of the state-of-the-art models available today. | One of the engineers on the launch announcement said, “Very soon, there will be no more benchmarks left,” implying that future versions of Grok will be able to master all of the benchmark tests. | | Source: xAI | One impressive feature about yesterday’s announcement was that it was completely unscripted. The team knew exactly what they wanted to share and were willing to run the model in real-time to demonstrate its capabilities. | While the team was presenting, they gave Grok 3 a task to accurately calculate a spacecraft mission from Earth to Mars and back. The task was also to create a graphic visual of the orbits of Earth, Mars, the Sun, and the spacecraft for this mission. | Below is a short video of the graphic of the mission that Grok created while the xAI team was making the announcement. | | Source: xAI | Grok 3 accurately calculated the mission and created the video visuals of the mission over roughly 10–15 minutes of “thinking.” And the team was able to show Grok 3’s chain of thought – its thinking process – as well as the code used to run the calculations. | Now, regular readers of The Bleeding Edge with a sharp eye may have noticed that OpenAI’s o3 model was not included in the benchmark analysis. That’s because o3 has not formally been released and is not generally available for use. | With that said, OpenAI’s o3 reportedly score 96.7% on the AIME math exam and 87.7% on the GPQA Diamond benchmark for PhD-level science questions, which would put it just ahead of Grok 3 on those benchmarks. | What does it all mean? | A Mission for the Universe | xAI did in seven months what it took others years to do. It has the leading frontier AI model that performs better than anything else generally available today. | And xAI’s approach is much like the approach that Musk and his teams take at his other companies. | Musk’s team iterates quickly and intends to improve Grok on a daily basis. Equally important is that Musk said that they would be updating Grok daily which each improvement of the model, something that no one else in the industry has been doing. All of the other major players have released a model, locked it down, and then continued work on a future model. | That’s not the case with Grok 3. We can expect improvements every day. I’ve been working with it for the last 12 hours. It’s phenomenal and has immense utility. For those curious, you can go to Grok.com and check it out for yourself. | The reality is that even if xAI updated as frequently as Tesla updates its self-driving software (which is about once a month), it would set xAI apart. Daily drops of updates make it all that more exciting. We can all stay on the bleeding edge of frontier AI models simply by using Grok. | And from what I’ve seen, I’d be willing to bet that Grok 3 overtakes OpenAI’s unreleased o3 model in a matter of weeks. | It’s also important to note that Grok 3 is a real-time AI model. It even has a DeepSearch mode which xAI considers to be a next-generation search engine. | This means that Grok 3 can search the internet in real time, combine that recent information with its learned knowledge, and produce very useful and comprehensive answers in a matter of seconds. | In other words: This isn’t just a race to AGI. It’s also a race to disrupt the hegemony of Alphabet with its Google search engine. | And it’s very clear that Grok 3 is already far better. Grok does in seconds what it would take minutes or even hours to accomplish using Google. And the results aren’t heavily biased like Google’s search results. | This is why xAI is racing ahead as quickly as it can. It’s also why xAI made a $97 billion offer to acquire the non-profit OpenAI. If successful, it would acquire widespread distribution. | Search – as a category – is already being redefined with these AI models. Alphabet wants to ensure it’s done using its own frontier model, Gemini 2. | Doing so will allow it to both control advertising revenues and data surveillance, as well as influence the general population in favor of its own political and industrial agendas. xAI and X, however, are looking to steal away market share with a superior product and a truth-seeking technology. | But as it is with Musk, the mission is always far grander than that… | xAI’s mission is to understand the universe. | And in Musk’s own words: | | In order to understand the nature of the universe, you must absolutely rigorously pursue truth or you will not understand the universe. You will be suffering from some amount of delusion or error. |
| | This is why what xAI is doing is so incredible. By optimizing for an evidence-based, factual general intelligence, it will – by definition – be optimizing for the highest-performance AGI. | What will happen in the months ahead will be profound. | I’ll do my best to keep you on top of these incredible developments, so we can grok what’s going on and what it all means for our future. | Jeff | | | | Like what you’re reading? Send your thoughts to feedback@brownstoneresearch.com. | | | Brownstone Research 1125 N Charles St, Baltimore, MD 21201 www.brownstoneresearch.com | | To ensure our emails continue reaching your inbox, please add our email address to your address book. This editorial email containing advertisements was sent to newsletter@newslettercollector.com because you subscribed to this service. To stop receiving these emails, click here. Brownstone Research welcomes your feedback and questions. But please note: The law prohibits us from giving personalized advice. To contact Customer Service, call toll free Domestic/International: 1-888-512-0726, Mon-Fri, 9am-7pm ET, or email us here. © 2025 Brownstone Research. All rights reserved. Any reproduction, copying, or redistribution of our content, in whole or in part, is prohibited without written permission from Brownstone Research. | Privacy Policy | Terms of Use | | |