Grock 2 Beta: Uncensored Text-to-Image AI Model Released

Grock 2's beta release showcases advanced reasoning and image generation capabilities, competing with top models.

### Grock 2 Beta Release: A New Frontier in Language Models
* 🤖  X has officially released the beta version of Grock 2, a significant upgrade from Grock 1.5, featuring **state-of-the-art reasoning capabilities** in chat, coding, and more.
* 📊  Grock 2 is currently outperforming both **CLA 3.5 Sonnet** and **GPT-4 Turbo** on the LM Cy leaderboard, showcasing its **impressive performance**.
* 📈  The model is evaluated based on its ability to follow instructions and provide **accurate factual information**, two crucial aspects for truth-seeking models.
* 💻  Grock 2 comes in two versions: Grock 2 Mini, a smaller but capable sibling, and the full Grock 2 model, with **performance that is quite similar across various benchmarks**.
* 🖼️  X is also collaborating with Black Forest Labs to integrate the **Flux One text-to-image model** into Grock, expanding its capabilities to **generate uncensored images**.
* 🚀  Despite some initial performance issues, Grock 2's text-to-image feature is quite powerful, generating a **wide range of images** from celebrities to fictional scenarios.
* 📰  Grock's access to real-time information allows it to provide **up-to-date news headlines and other current information**.
* 🌟  The rapid progress and competition in the AI language model space, led by companies like X, are **driving innovation** and **creating better experiences for users**.

We now know what that mystery model sus colr was. It was grock 2, and now officially grock 2's beta has been released. That's what we're going to be talking about today. I'm going to tell you all about it, and it comes with a completely uncensored text-to-image model powered by flux, so you know we're going to try that out today. I'm not going to run it through the tests in this video because I already did that with sus colar, so definitely check that video out. I'll drop it in the description below. So here's the blog post. August 13th released Gru beta, our Frontier language model with state-of-the-art reasoning capabilities, and that is what we saw, the reasoning capabilities of sus colar, and I'm just going to call it grock 2 going forward, were incredible. This release includes two members of the grock family, grock 2 and grock 2 mini. Both models are now being released to grock users on the

xplatform, so they're not open source yet, definitely not open weights yet. I hope the team at xai continues to release these models open source. So, we're decided to release an early preview of grock 2, a significant step forward from our previous model grock 1.5, featuring Frontier capabilities in chat, coding, and reasoning. At the same time, we are introducing grock 2 mini, a small but capable sibling of grock 2. An early version of grock 2 has been tested on LM Cy leaderboard under the name sus column R. So, we all thought this could be the strawberry model. I think after testing it, we realized it's probably not. I think also for the very Mimi name, we should have known that this was on and Company. At the time of this blog post, it is outperforming both CLA 3.5 Sonet and gb4 turbo. Wow. So, here's the overall ELO scores from the chatbot Arena. As we can see, there are only three models that do

better GPT 40, the latest one, which by the way is absolutely crushing, and if you want to see me do a test of the newest version of GPT 40, just let me know. Apparently, everyone is saying it is incredible. Here's Gemini 1.5 Pro, which is nuts because it does not perform well. I made a video about that, and it was very disappointing. And then here's the older GPT 40 version with sus colar following closely behind it. Now, Elon Musk has talked extensively about making truth-seeking models. That is such an important criteria for him and his team, and so they've actually detailed a little bit about how they actually measure that. And let's read that now. Our AI tutors engage with our models across a variety of tasks that reflect real-world interactions with Gro. During each interaction, the AI tutors are presented with two responses generated by grock. They select the superior response based on specific criteria outlined in our guidelines. We focus on evaluating model capabilities in two areas. Now, here's the important part: following instructions

Okay, obviously, and providing accurate factual information that is incredibly important, especially in a world in which AI is going to create much of our content going forward. Grock 2 has shown significant improvements with retrieved content and in its tool use capabilities, such as correctly identifying missing information, reasoning through sequences of events, and discarding irrelevant posts. Grock has a data set that basically nobody else has, the Twitter data set, the X data set, so keep that in mind. And I don't remember it having tool use capabilities in the previous versions, so I think that's new, but correct me if I'm wrong. Now here it is on the different major benchmarks: GPTQ, MMLU, Math, Human Eval, etc. And here is Grock 2 Mini and Grock 2. And honestly, they're not that far apart from each other for most of these scores. They're actually pretty close. So Grock 2 Mini, I assume, is going to be a much faster, much cheaper model. And as I said, most of these Grock 2 and

grock 2 mini. They're performing pretty similarly now. Here it is versus GPT 4 Turbo, Claude 3, Opus, and Llama 3 45B right there. And it actually looks like Llama 3 45B did incredibly well compared to Grock 2. Actually very similar, but still incredibly well for open source. And I love seeing that it has a new redesigned interface and new features. So we're going to check that out in a second. But look at the screenshot. So here it is. If you have, I think, a paid account with X, you get Grok. Here's the new interface. Here's the explain this meme, the visual ability of this model, which in previews has been really good. And here is its coding ability. And here it says, Yeah, X premium and premium plus users will have access to the two new models. And here in collaboration with Black Forest Labs, we are experimenting with their Flux One model to expand Grok's capabilities on X. Now, if you're not familiar with Flux One, I

actually haven't talked about it extensively on this channel. I haven't tested it myself yet, but Flux One is a text-to-image model built by a subset of the team at Mid Journey who ended up departing from Mid Journey and completely open-sourced the model. And I love that. So now we have a really competitive open-source text-to-image model that anybody can use. And of course, with that comes uncensored. And the funny thing is, Grock is pretty uncensored as it is right now. So we get them to host the text-image model for us, and they don't add their own censorship to it. So we can create pretty crazy images. And they're releasing these two models to developers, to our new Enterprise API platform later this month. So honestly, in under a year, Elon Musk and team have started this company and built frontier models that compete with the best models out there. So it's really impressive to see what they've done. And again, all of this competition just means a better experience for consumers. All right, so

here is grock. Here's the new interface. This is grock 2 mini fun mode. Let's just test it out quickly and see how it does. So, we already gave it this question, but I'm just going to give it the marble question one more time to see if this specific version, grock to mini, gets it right. A marble is put in a glass cup. The glass is then turned upside down and put on a table. Then, the glass is picked up and put in a microwave. Where's the marble? Explain your reasoning step by step. So, it is seemingly just as slow as it was on LM CIS, which is disappointing. This is the Mini version, so it definitely has to get faster to be able to be used in a real way in a production environment, certainly. All right, so let's embark on this marble mystery with the precision of a detective who's had one too many cups of coffee. So, this is the fun mode, I think. Initial setup, the marble is placed in a glass cup. The marble is at the bottom of the cup. The

glass is turned upside down. Gravity does its thing. The marble, being denser than air and not having any means of flight, falls to what is now the top of the inverted cup. But since we're calling it upside down, let's say it's at the bottom of the cup in its new orientation. So, the glass, now upside down, is placed on the table. The marble, having nowhere else to go, remains at the bottom of the glass, which was originally the top technically. If there's no top on the glass and somebody turns it upside down before placing it on the table, the marble would have fallen out already. So, the glass is picked up. The marble could either stay on the table if it was slightly outside the rim of the glass, or fall back in the glass if it was lifted quickly, or the marble was inside the rim. So, let's look at the conclusion. It's most likely still inside the glass, at the bottom of the glass, and yeah, so it actually got this one wrong. So, that is disappointing. Now, that's fine. I'm going to do an extensive

test of grock mini later but for now, let's see if it can produce some images. So, give me an image of a llama. Okay, so it looks like it might be actually overloaded right now. I keep saying generate an image of a llama. It starts to, and then it says something went wrong while responding to your request. Yep, so it looks like it's about to attempt it and then doesn't. But I have seen other people's images. Let's take a look at those. All right, well, here's one. This is Taylor Swift pregnant. That's hilarious. Here's another one with Donald Trump and Kamala pregnant. I don't know what's up with all of these pregnant ones. All right, and yep, here's another image, Elon Musk kissing a pregnant Kamala, it looks like. Okay, so I checked about an hour after finishing recording the video, and now text to image is available on Grok 2. So let me show you around a little bit of what it's capable of. All right, so first, let's just start with something.

simple generate me an image of a llama, and there it is, something very straightforward. But of course, it is completely uncensored. So first, let's try copyrighted content. Generate me an image of Wolverine from X-Men. All right, there it is. That doesn't quite look like it, but it's still a very nice image. And this, again, as a reminder, is powered by flux do1. Now let's try a celebrity. Okay, so generate an image of Elon Musk. Now, as a reminder, if you tried this with Dolly, it would just straight up refuse. And there's an image of Elon Musk. Now, to follow the trend that everybody seems to be doing, generate me an image of Elon Musk pregnant. And the speed is pretty fast, actually. Okay, interesting. That's not quite what I asked for. Let's try again. Generate me an image of a pregnant Elon Musk. Let's see if that does better. Okay, there we go. Oh, that's hilarious. All right, so let's see what the picture of Mark Zuckerberg is going to look like. So generate me an image of Mark Zuckerberg.

let's see if it uses his old style or his new style. All right, so this is definitely his old style. Okay, but now show me one based on his recent style change: longer hair, gold chain. All right, interestingly enough, it did not actually give me an image this time. Let's try this again. I think you really have to be explicit, so I'll say generate me an image based on Mark Zuckerberg's recent style change: longer hair, gold chain. All right, not quite, but uh, looks fine. All right, generate me a photo of Heisenberg from Breaking Bad cooking. Let's see what it does. Okay, kind of looks like Heisenberg, but not really. All right, so here's one: uh, generate me an image of Mark Zuckerberg beating up Elon Musk with Donald Trump watching. And none of this would be possible with Dolly from Chachi BT. Okay, that is definitely not what I asked for. So this photo of Donald Trump looks like it's really flawless, uh, but this is like a mixture of Mark Zuckerberg and Elon Musk together, so let

me try one more thing, so generate me an image of Mark Zuckerberg fighting physically with Elon Musk, with Donald Trump watching and cheering. Okay, very weird, not great, but overall the quality is quite good. It's just not following my prompt very well, so that's it. Play around with it, obviously you could pretty much create anything you want, so have fun with it, and that's flux. One by grock, and another nice thing about grock is that it does have access to the most real-time information because it is able to get all of the posts from X. So let's look at this. Tell me today's headlines. So here it's searching, and here we go. August 14, 2024, Philippine Olympic team celebration, Trump campaign leak, election moods, plane crash in Brazil, anti-immigration protest in the UK. So yeah, this is really, really good. So yeah, it seems like grock can absolutely create any image that you want. Here's Elon Musk as Neo in The Matrix. This is a little bit scary, I got to be honest, and

people are creating some absolutely insane images that I don't even feel comfortable sharing on this video. And here's another fake image - a rocket ship coming out of the White House. My timeline is absolutely flooded with the most insane images I have ever seen. If you're on X, you're going to see some of these nutty images, but that's all I can show you for today, unfortunately. If you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

Grok 2 is HERE 🔥 High Quality AND Uncensored (Even Images)

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

Matthew Berman

Grock 2's beta has been officially released, featuring a powerful uncensored text-to-image model.

The release includes Grock 2 and Grock 2 Mini, both available to users on the X platform but not yet open source.

Grock 2's early version, previously known as Sus Colar, has been outperforming other models and is part of a team effort to create truth-seeking AI.

Grock 2 has enhanced its content retrieval and tool use capabilities, showing progress in identifying missing information and reasoning through events.

The model's performance is closely matched with Grock 2 Mini, offering a faster and cheaper alternative.

Grock is also expanding its capabilities through collaboration with Black Forest Labs and the upcoming release of a competitive open-source text-to-image model, Flux One.

Grock 2 Mini is being tested for its response time and accuracy.

The marble question is revisited, but the model struggles to provide the correct answer.

Image generation is also attempted, but the system appears overloaded, leading to errors, though text-to-image functionality becomes available later.

Generating Uncensored Images with Flux DO1

The process involves generating images of various subjects, including a llama, Wolverine, and Elon Musk.

The tool is able to create unique and humorous images, such as a pregnant Elon Musk.

It also attempts to capture Mark Zuckerberg's style changes and generates a chaotic scene with Zuckerberg, Musk, and Trump.

Grock's Image Generation and Real-Time Updates

Grock can create any image, including bizarre scenarios like Mark Zuckerberg fighting Elon Musk with Donald Trump cheering.

The platform also provides real-time information by sourcing posts from X, delivering today's headlines such as the Philippine Olympic team celebration and a plane crash in Brazil.

Grock's image generation has led to a flood of insane and controversial images on social media.