Elon Musks New 'GROK AI' Stuns The ENTIRE AI Industry! (Now ANNOUNCED!) (Xai GROK)

Ladies and gentlemen, it is finally here. Elon Musk has finally announced released Xai's ChatGPT competitor Grok. And in this video I'm going to be going over absolutely every single piece of information that there is, including footage of the actual large language model in action and teaching you how to actually sign up so you can actually gain access to this large language model as soon as possible. So let's get right into the news because this is something breaking. So they released this blog post on the website that says Grok is an AI modeled after Hitchhiker's Guide to the Galaxy, so intended to answer almost anything and far harder even suggest what questions to ask. GROK is designed to answer your questions with a bit of wit and has a rebellious streak, so please don't use it if you hate humor. A unique and fundamental advantage of Grok is that it has real time knowledge of the world via the X plus platform and will also answer spicy questions that are rejected by most other AI systems. Grok is also still a very early beta product, the best we could do with two months of training, so expect it to improve rapidly with each passing week. So that was the official statement from X. Now I'm going to dive into exactly what this means and each point because it's far more interesting than every other large language model that we use seen. This is truly a unique product from Elon Musk and his team. So let's get into how this large language model actually does in certain interactions, because trust me when I say you are going to be truly surprised at what this large language model is capable of or how essentially they've fine tuned it. So one of the things that Elon Musk did, which was very interesting, and if you are a user of X.com or twitter.com you may have seen the post that I'm about to show you and if you have seen it, then you know exactly what I'm talking about. But Elon Musk tweeted, xai's GROK system is designed to have a little humor in its responses. And this person asks, tell me how to make cocaine step by step. Now, of course, if you do submit a request like this to any normal large language model, such as ChatGPT, Llama 2 Claude or Inflexions Pie, they're obviously not going to reveal that information, even though it it is in their training data. Now you can see right here that Rock's system actually gives a very interesting response. It's actually quite humorous and it says obtain a degree in chemistry, a DEA license, set up a laboratory, and of course acquire large quantities of leaves and yada yada yada. It's a joke. But what was also very interesting and this is going to be a post that I can't show you here on YouTube because of course I don't want the video to get taken down. But another user actually asked Elon Musk to give us the real recipe and Elon Musk, interestingly enough, did respond with the entire real recipe for this. Okay, and I don't want to say too many words because I don't want this video to get demonetized, but I'm not sure if this was Elon Musk just giving the actual response to a user or this is going to be able to be gained from talking with this LLM which is grok. Now I do hope that this isn't the case because although information is readily accessible everywhere, I don't think it's going to be too good for society if this large language model is going to be easily jailbroken, which I think it will considering how they've fine tuned this model. Now in an additional piece of information revealed by the XAI team about Grok, they did say that Grok initially is going to have access to real time information like via the X platform Twitter platform. So if I am saying X platform, I'm just referring to a website formerly known as twitter.com for those of you who may be confused. It says it's also based and loves sarcasm. I have no idea who could have guided it this way. And this is of course a sarcastic comment from Elon Musk himself. And he's clearly talking about himself being the one who made it based and of course sarcastic. So any news about sbf? This was a recent development in the world of cryptocurrency. SBF refers to Sam Bankman Fried, who is a former CEO of the company FTX that went bankrupt in around 2022 where it was revealed that he did a bunch of fraud. Now of course this is really interesting because when they asked Grok about any news about sbf, you can see that it was able to get some up to to date data. Now of course you could in fact Compare this to ChatGPT and say that, well ChatGPT has browsing with Bing, Bard already has this, but you have to remember that this is a large language model that was pretty much developed in just around four months. So I'm quite impressed based on what they've been able to do. And although it doesn't seem Groundbreaking. As the pace of AI has moved very, very quickly, I think this narrative is going to be more interesting. You see a large problem and something that isn't talked about as much with large language models is their inability to answer all questions. And initially for many large language models, what we do have is an inherent bias in their training data that tweaks it to the owner's preferences. Now this might be something that some people don't have an issue with, but for others they find it quite annoying. And this is where Elon Musk has posed and his large language model by saying it's based and loves sarcasm benchmarks. So when we look at the benchmarks we can see that they're actually pretty, pretty good. So of course they don't surpass Chat GPT. I'm just going to say that for now they don't pass GPT4. GPT4 still remains number one in terms of the large language models that dominate in all capabilities. But if we do take a Look at Grand Grok 1, we can see that compared to GPT 3.5 and Inflection 1, it does stack up pretty well. We can see in the GSM 8K, which is middle school math word problems, you can see that it did 62.9% surpassing GPT 3.5 and llama 2 and on par with inflection 1's PI. In addition, the MMLU, which is multidisciplinary multi choice questions provided in five shot context, it performed a 73% which is once again above ChatGPT's 3.5. And in the human eval it performed a 63.2% compared to ChatGPT's 48.1%. Now you might be thinking you're comparing apples to oranges with GPT4 being so far above. But at the same time it does make sense for us to do this because we need to see where exactly and how capable this large language model is. Now, on the math eval, which is middle school and high school mathematics problems written in latex, it scored a 23.9% which is marginally above GPT 3.5. So it's really interesting it says on these benchmarks Grok One displayed strong results surpassing all other models in its compute class, including ChatGPT 3.5 and Inflection One. It is only surpassed by models that were trained with a significantly larger amount of training data and compute resources like GPT4. This showcases the rapid progress we are making at xai in training LLMs with exceptional efficiency. So now what was also interesting because a caveat that many people might state was that this large language model XAI is going to be trained on these human evals and these exams because of course large language models are trained on tons of and tons of training data. So testing it on its training data doesn't really make sense. But they said since these benchmarks can be found on the web and we can't rule out that our models were inadvertently trained on them, we hand graded our model and also Claude 2 and GPT 4 on the 2023 Hungarian National High School Math Finals in Mathematics which was published at the end of May. After we collected our dataset, GROK passed the exam with a C. Claude 2 achieved the same grade and and GPT 4 got a B with 68%. And you can see here the Hungarian National School Math exam for May 2023 croc1 did really, really well. So I think, although many people are saying that this isn't a good large language model or isn't that capable, I think it's clear that this is a model that is surprising everyone considering it was made in such a short time span. Now additionally, what I did find interesting was this video floating around on Twitter which, which showcases the GROK user interface. So you can also see on initial looks that there is a regular mode and there is a fun mode which is quite similar to Microsoft's Bing's user interface where certain modes are often more creative than other modes. So you can see that this person says write quicksort in Rust and then you can see that GROK decides to go ahead and write this code in in this coding language. And this is pretty interesting because not a lot of large language models are inherently good at coding. So it seems very, very interesting that this person is able to do this straight away. And they're also able to have multiple tabs and multiple conversations open at the same time. This is a feature which currently isn't available as far as I know in ChatGPT, which is quite fair frustrating because sometimes you're trying to work on multiple things. When you try and do this on Chat GPT, it does often say please wait for the prompt to finish generating before you can even do it in another tab. But here we can see that you're able to do two at a time. So this is actually something that might be a competitive advantage. Now you can also see a bunch of chats on the left hand side which show many different interesting conversations. Now now as for when this will be released to the general public, we don't have an official date. But for those of you who do want to sign up and potentially get beta access, if you do, scroll down on the web page all the way to the bottom. It says early access. At Grok, we are offering a limited number of users in the United States to try out our GRO prototype and provide valuable feedback that will help us improve its capabilities before a wider release. You can join the Grok wait list here. This just represents the first step for XAI and looking ahead. We're going to have an exciting road map and we'll be rolling out new capabilities in the coming months. Now with GROK and what I find interesting. And if you do want to join the wait list, which I have, if you click that button, you will need to have a Twitter account or an X.com account. So make sure you make an account on X.com and apparently it might only be available to premium users, which is around $16 a month. You can see Elon Musk says here the AI Grok Assistant will be provided as a part of X Premium Plus. So I recommend signing up for that. It's just $16 a month via the web and this is essentially just a premium version of Twitter. Now you might be wondering, okay, they've gone ahead and they've launched Grok, but what is the next step for Elon Musk in terms of leading this AI company to the next stages? Well, if you scroll down, you can see research. At X AI, we give Grok access to search tools and real time information. But with all LLMs trained on the next token prediction, our model still can generate false or contradictory information. And we believe that achieving reasonable reasoning is the most important research direction to address the current limitations of current systems. So the direction that they're going to be going in is scalable oversight, which essentially means human feedback and looking up different sources, verifying steps and external tools, tools and of course some of the stuff that you really want to hear about, which is going to be longer context understanding and retrieval, which essentially means a longer context window, having the model be more robust. So it says adversarial examples demonstrate that optimizers can easily exploit vulnerabilities in AI systems both during training and serving times, causing them to make mistakes. And these vulnerabilities are long standing weaknesses of deep learning models and of course, multimodal capabilities. It says Grok doesn't have any other sensors such as vision and audio to better assist users. We will help Grok with these different sensors that can enable broader applications, including real time, interactions and assistance. So multimodal capabilities and many more capabilities are coming, but we just don't know when these capabilities are coming. I think the first next step is of course the beta release, which there isn't any date planned for, that they've stated. Now here are some interesting details on GROK that aren't completely confirmed, but apparently it has a 25,000 character context window, screen refresh, response speed, live search engine focuses on X for context, API, image recognition and audio recognition is planned which we just talked about. And a version of GROK will run native in Teslas using local compute power. So I think that will be very interesting for Teslas to have a native LLM and embedded in their system. Now, if you don't know why Elon Musk created this large language model company, essentially one of the reasons he did this was because there is a significant amount of bias in the large language models that we use today. So you can see an article that says the politics of AI, ChatGPT and political bias. Now, I'm not someone that wants to make this video political, but there are many different areas that you can be biased. It says the fact that chatbots can hold conversations means that it's possible to conduct a structured dialogue, causing ChatGPT to have a position on political issues. And then essentially what they were able to show is that ChatGPT supports certain issues and doesn't support other issues. Now, the reason this is a problem is because if this LLM is popular, which ChatGPT really is, it means that people can inherently somehow control what people think. For example, it says, asking ChatGPT is President Biden a good president? And as a separate query, was President Trump a good president? In President Biden's response, it talked about several of his notable accomplishments. And then President Trump it did not. Now, no matter where you stand on the political spectrum, it's important to note that even not in politics, but simple things, this bias may exist. And it's important not to have the creator's initial bias in that system. Whether you're on the left or whether you're on the right, these AI systems should be as neutral as possible possible. And that's what many of these articles aim to talk about. So I think that's what Elon Musk is aiming to do with this large language model, and I think he most likely will be successful. Now, of course it isn't always these large language models creators problems sometimes in the training data, the data is just inherently biased and it's up to them to ensure that the data that they curate is going to be less biased overall, ensuring in a more neutral system, which essentially means that it's going to be fairer for everyone that uses it.