Grok 3 is Here - Smartest AI on Earth?

I finally got access to grock 3 so I wanted to show you exactly what it could do in this video they have some really useful options with grock 3 like deep search and a reasoning model that I wanted to show you it's currently available inside of x. com but you do need a premium subscription so you do have to pay for it in order to get access to it once you do pay for that subscription you'll see this little grock icon and here's the pricing right now so for the monthly fee of $8 you will get Gro and this is the one I have right now which removes ads and some other things that come with x. com and you could also use grock 3 at gro. com again you do need to sign up for the premium and use that to log into an account here and then you'll have access to grock 3 over here with deep search and reasoning model too now just to give you a quick background on grock 3 this is obviously the third version of grock which is the large language model from X aai Elon musk's AI company and the reason they were able to catch up so fast is they built this data center here in just 122 days they were able to get 100,000 gpus to fully train grock 3 they've recently expanded that to 200,000 gpus currently this is the largest cluster of gpus to train large language models in the world and earlier chatbot Arena which is a website where you'll see two answers from two different models and choose the one which you prefer and not see which model it is before you vote grock 3 got the highest score here it beat Gemini it beat deeps R1 01 preview the latest 01 it beat every single model here but we're going to go ahead and test that out for ourselves okay the very first thing I want to show you is it's writing style by default without doing anything to The Prompt so getting a 500w blog post about the release of grock 3 and by default this has access to the entire x. com or the old Twitter database so it will search that and it will search the web you don't even have to turn on any kind of search function that is already its default anytime you're doing any kind of research grock 3 is here x ai's latest sleap in AI awesomeness so it definitely has a more quirky and unique kind of tone compared to any other chatbot I've used big news tech fans as of February 19 2025 xai has officially unleash grock 3 into the wild that's the default you could always tweak that with how you give a prompt you could ask for a different tone or a different writing style or a different reading level but this is kind of what you're going to get out of default totally different sense of humor than chat GPT gemini or really any other model out there right now oh this is interesting it says word count 500 on the dot okay let's find out if that's true word count is 513 wow that's actually shockingly close to what I asked for they usually don't get it anywhere near that kind of word count now this has also multimodal capabilities so not only could it see inside of images with his vision capabilities it could also generate images so let me attach an image here and then we'll go ahead and generate one too I'm going to actually attach that picture of that data center I showed you and I'm going to ask it what it is it has text it has an image here so the image appears to be an aerial view of the largest Data Center and super Computing facility label as Colossus 200k GPU that's correct okay so that did its job but one thing I've noticed lately is ever since I started using it I've been seeing this message sometimes we're seeing some heavy traffic so we've opted to an alternative model to get you answers faster let me just see what does this mean okay right here it says the system switches when it has a lot of traffic instead of using grock 3 uses a less resource intensive model so that might be another model they actually have that is not available here for picking called grock 3 mini which is going to be rolling out as part of their API soon too okay let's see how well it follows up with a previous conversation so just in this conversation I asked how much did it cost to build colaus so that image kind of told us how many gpus were used Colossus initially launched with 100,000 Nvidia h100 gpus in July 2024 by February 25 expanded to 200,000 gpus see very up todate because it not only uses web page search again without turning anything on two of my favorite things here is web search is always on if he needs it and I don't have to choose from a 100 different models on top like every other chatbot is trying to do now chat GPT is trying to move away from that when GPT 5 comes out is supposed to be the only one you get to choose and then it'll figure out which one to choose in the background to give you the best answer this still has deep search and thinking that I have to turn on but at least for the search part I don't have to pick anything and

estimated cost these gpus are typically somewhere between like 25,000 to 60,000 says 5 to 6 billion in GPU Hardware 3 to 5 billion in Cooling and upgrades and things like that construction and it's saying by the time they expand out to 1 million gpus it' be somewhere in the $30 billion range and it could also create images so we'll go ahead and test that out just to show you a couple of different examples create a picture of a dog running on their moonlight okay that's not too bad let's click on a different one okay that's pretty good too and you could always edit these two and use a text prompt here to make subtle changes to this image too and then download it from here okay the next thing I want to show you is content creation specifically for x. com because what better way to post on here than use grock to figure out what actually works give me the top 10 posts about the release of grock 3 now this is going to search the web and other people's posts so you can see all the different post that is pulling from you could quickly see things this way and I could do a simple follow-up analyze those 10 tweets for me and give me a post that is educational because that's the type of content I post and here's the post that he wrote for me it's okay I probably needed to give it a little bit more instruction for his tone but by default analyzing what's there already a lot of them were in this kind of tone XIs Powerhouse AI is redefining speed and smarts in just two days right things like that I wouldn't about but the first part of that is actually pretty good and I could ask for 10 more like it but less Promotional and these are a lot more like things that I would want to post now next I want to show you deep search which pretty much every app now chat jpt has this Gemini Google Gemini has this I'm going to use this image here from the noron newsletter which says they just got acquired this is from today so I going to see what's going on here to see if he understands that part okay the main text here we got acquired suggest that neuron has been purchased or taken over by another company okay so you got the gist here now we're going to turn on deep search and let's see if we figure out if I give it some context about the neuron to figure out how much things like this sell for and I'm just going to tell it the neuron's been around a couple years in the AI space has 500,000 subscribers let's see what it comes up with and deep research actually has a whole different kind of graphic here how long it's been thinking so far it's going to go through a bunch of different links right and it's going to find out here is a good example morning Brew which was a newsletter I acquired for $75 million in 2020 so it's finding all this information here okay this took only 52 seconds went through 90 different sources and I've used the Google one they take much longer 52 seconds is actually really short typically they take six seven sometimes even over 10 minutes to do this and 83 web pages right on top around $18 million wow I definitely don't think that's true let's see how he came up with that $18 million because morning Bruce sold for 75 with 2. 5 million subscribers The Hustle sold for 27 with 1. 5 million subscribers we estimate the neuron annual revenue at 3. 5 million using Revenue per subscribers Benchmark of the morning Brew $8 per subscriber and the hustle $667 okay interesting really interesting but hey we'll never know this is not my company we have a large newsletter too it's not quite at half a million yet but interesting numbers okay now I'm going to see how it deals with any kind of data file so I have this Amazon CSV file this is just a data document about Amazon stock price I'm just going to ask it what this is first based on the data range the low initial price and the trading volume this data set resembles the historical stock price of Tesla that is not correct literally the file says Amazon on it so you got that wrong okay I told it it's wrong it's Amazon and I want to see if it could create a graph for us which we could do inside of chat jpt it's asking me if I want to generate it I'm going to say generate it okay so yeah it definitely can't create a graph it created this instead and chat GPT a whole lot more useful for kind of breaking down data into visual Graphics Claude is also really good at that so I use Claud for that often as well okay here I have a reasoning test and I tested this in my last video where I compared the reasoning model of deep seek I compared chat gpt's reasoning model and Gemini's reasoning model and now we have a grock reasoning model you have a rope that is exactly 50 ft and the building that's 75 ft you need to measure the height of the building using only the rope and your own body no other tools your 5T how can you do this described step by step R1 deeps car1 got a right Gemini got a right chaty PT did not get a right and I use gpt3 mini for

that test okay so the thinking graphic looks kind of different and you kind of see it's reasoning here so sometimes chat GPT hides that or gives you a little summary you can't really see it R1 breaks it down in very clear detail of how he's thinking through it so I'll let this finish up here and it is showing you how many seconds he's thinking for as well okay it took a 118 seconds and if I extend this out here you could actually go through the entire process of how it was trying to think through this was probably the longest any of the models that I've tested out had to think through a problem here and the answer that he gave us is probably the most wrong answer out of any of the models even the chat GPT model is telling me to put the Rope at the base of the building and somehow extend it out to 50 ft up in the air not how the physics of a rope works and then climb that rope and then use my body for the rest of the 25 ft that's missing where the Rope cannot measure it so okay I'm going to tell that it's completely wrong there is something called gravity where I can just throw the rope up there let's see if it kind of tries to fix itself okay now it's giving us a different way to do it this time we're going to go to the top of the building we're going to drop the Rope down from the top it's going to reach 50 feet so there's going to be 25 ft left from the bottom and then it says you might visually estimate this will use your own body as a measuring stick well that does doesn't quite work that's not going to be very accurate we're just kind of eyeballing it then now Google and deep seek gave me an answer with similar triangles that makes a whole lot more sense than kind of Standing On Top of yourself or dropping down from a 50ft rope from the top of the building oh right here it says you have one last remaining question for grock 3 with the think option turned on before the limit resets I literally maybe used it like six times here six questions with this turn on wow that is a very small limit I couldn't even really get past one reasoning question but yeah it did agree with me that similar triangle technique will solve that problem so I'll just try a quick multiple choice question this one every model I tested out got a different answer okay so it took 87 seconds this time going through it here and let's see what we got on the very bottom oh it's still going here oh wow it did get it right the answer is B so only chance gpt3 mini got this right the first time I tested a deep seek R1 and Gemini both could not get this right so let's say you're paying for chant GPT is it worth paying for grock in addition or instead and I think right now unless you're using this for its tone and you don't really want to train chat GPT to write the same way or if you really want to post on Twitter or on x. com and you want this to utilize that kind of analysis that it has well worth it if you want really up toate information this is probably the closest you're going to get to anything to real time because it has people's posts now those are just random people right so if you want more reputable sources it does also search the web so you could do your research that way but right now it's just missing way too many practical things for me for example it doesn't have system instructions I can't tell it to write in a very specific way chat GPT has that Claud has that and I can't add that here unless I put it inside of my prompts the way it analyzed documents I was not that impressed with it's deep search I think it's pretty good I got decent results a few times I tested it but I still think I'm getting better ones out of Gemini when it comes to deep research and obviously there are some big things that I use inside of chat GPT like those custom gpts in Claude there is something called projects that I use all the time in Google we have a huge context window so I could give it A500 page book so I feel like there's still some things that this could improve on to catch up with those other things but on the large language model side the speed side of things the reasoning side is definitely come a super long way in a very short amount of time so pretty impressive I think it's kind of worth a try but I want to switch from something I'm already using I might want to add it if I see some of those use cases for your practical applications or for your social media posting if you haven't watched the other video where I compared the other reasoning models I'll post that here that one is very deep dive Compares them in detail and kind of scores them at the end too thanks for watching I'll see you next time