ONE MONTH LEFT! New MAJOR Robotics/AI Breakthrough, ChatGPT Loses Its Mind, Google Gemma, Major AI

An insane week in AI. It's important that we make sure we stay on top of the AI news. And that's what this video is going to show you. Everything you missed. And apparently this first piece of news shows us that it might just be one month until everything changes. And if you don't believe me, take a look at this. So there was a recent tweet by Ted and this is someone that actually works at Google AI, Google DeepMind. And essentially this person is tweeting about robotics and scaling large models and, and essentially he is someone that is working in the field of robotics. Now this tweet that you can all see right here, if I zoom in, he stated that there will be a three or massive pieces of news coming out in the next weeks that will rock the robotics and AI space. Adjust your timelines. It will be a crazy 2024. Now of course underneath that he does state that it is not AGI and some people are stating that vague posting is not permitted. But I did find it quite fascinating that he wasn't the only person that stated that this will be crazy. I'm going to come back to this tweet because despite from him, someone else is also getting, I'm guessing whatever is going on in the robotics community potentially. It seems like there has been some kind of internal thing that's been found within the robotics community. I don't pay that much attention to robotics, but I will be of course, due to the future. But Berndt also said, and this is the guy who is the CEO and the founder of, of essentially 1x Robotics and this is the company that is working on full on end to end autonomy, essentially stated that new progress update on the droids dropping in four weeks. Looks like Moravik's paradox might be false and we just didn't have the data. And if you don't know what moravc's paradox is, it's essentially the theory that what is difficult for robots is really easy for humans. And what is of course easy for robots is really difficult for humans. So you know how, you know, AI systems and robots like, they can like literally just do maths really quickly. And for humans sometimes we do struggle, but for humans we have great dexterity. We can grab things like an apple, we can peel a banana and it, you know, literally we can run off like a Dorito chip for an entire day. That is not something that robotics can do. So it seems that maybe there might be some kind of breakthrough that's going on. Now what's crazy about this is that of course, with burnt and 1x, they seem to have already made some kind of breakthrough. Because what we're seeing right here in this video, we're seeing where we see these robots doing this end to end autonomy. This is one of the first kind of demos in which we've seen recently the kind of robotics being able to do this at such a really, really decent rate. So it seems that whatever kind of barrier that was stopping them previously, maybe they've made that. And it seems that maybe across the board it has been shared with these teams and it seems like since the timelines, because this guy said dropping in for weeks and then of course someone said that there is a massive update coming out in the next couple of weeks that will rock the robotics AI space. I think it could be something really, really crazy. Maybe it's going to be a new architecture that combines the recent technology that we've seen with the robotics architecture. Maybe it's going to be something different. I honestly have no idea. But what I do know is that someone saying that adjust your timelines. I feel like we've already adjusted our timelines with what we've seen with Sora because we kind of didn't expect that technology to be that good. But this isn't someone that is. I wish, I guess you could say farming engagement, because sometimes there are people out there that do farm engagement and do farm. I guess you could say a lot of hype, but I don't think this is someone that is doing that. But it does seem like across the board we are going to be getting some kind of massive update. So I'm going to be excited for that. But at the same time I am going to be a little bit apprehensive to this. And I'm not someone who's a doomer or like a Luddite or someone that just is afraid of technology. But I think the problem is, but I think the biggest concern, I guess you could say, is that I just don't know what's coming. Like nobody knows what's coming. A lot of the researchers don't know what's coming. We don't know how society is going, going to adapt. We don't know what's going to happen with ubi. We don't know if there's going to be, you know, riots in the streets for people, you know, going back. I mean, it's, it's just a very, very uncertain time, which is, I guess I could say a little bit concerning. But at the same time it's also very, very Exciting, because the potential for good and the potential for abundance is really, really here. And this is something that we just didn't think would happen potentially even in our lifetime. So it's an exciting period to be in and also a quite daunting one, depending on where you stand on the spectrum. So that kind of update is really, really important. If you did miss, yesterday's video is likely one of the most important videos. I'm not going to say it's going to be the most important video of the year because we've got a long year to go, we've got a long, you know, 10 months. And of course we don't have no idea what OpenAI and Google is still kicking up. But essentially there was a magic breakthrough that essentially got this person to invest $100 million into a AI coding startup. And it was pretty crazy because this Guy essentially wrote 100 million check to the developer of an AI company and essentially not the developer to the CEO, of course, because essentially they saw something that apparently is just so good that they just essentially just had to invest straight away. And I think what they essentially described it was, was they essentially described it as something that had active reasoning capabilities similar to the Q Star model developed by OpenAI last year. Now, I'm guessing that most people did watch the video yesterday in which I talked about how this is pretty much it, because now there's kind of this race to the bottom. And I want to a point, when I said race to the bottom, a lot of people saying, why is it a race to the bottom? It's a race to the bottom because it's something that, you know, of course technology gets better, we get better technology, of course we get better products, of course that is good. But on the other side, you do have the existential crisis, which is where, you know, deep fakes get worse. We do have more misinformation campaigns, we do have better image generation, which is of course, you know, bad for, like I said, misinformation. AI takes the video. If that accelerates on a crazy timeline, we of course have even more information misinformation. And of course we have the existential crisis, which is super intelligence being developed, which is exactly what this company is doing. And if they do that, we could have some kind of, I guess you could say, existential threat being developed. Which is why I say that it's a race to the bottom, because whilst you're developing this technology, you do have to be very, very safe. But if you're not being safe, it's a Race to the bottom. Because if they just put out a super intelligent entity or an agile system and people manage to jailbreak it and they're able to just cause havoc in ridiculous capabilities, it's not going to be good for anyone. So that's why I say there's a race to the bottom. Because whilst these systems are good, we also always do need to ensure the safety of the public. And that is something that I think that when two companies are fighting to essentially have the spotlight and be the best company and of course deliver a return to their investors, it can produce a negative outcome. And this is something that we talked about with Moloch, which is essentially where it's an unintended consequence or of your competition doing something that you wouldn't usually do. But if they don't do it, you lose the business and so you have to do it and then everything just gets worse. Okay, so essentially it's going to be pretty crazy because if you don't know what Q Star was, Q Star was a real breakthrough that shook OpenAI to the point where I think even Sam Altman was fired. Ilya tried to get him removed from the board. There was this whole, whole host of things. So this article is really, really important because these guys not only said that they made some kind of breakthrough which could enable active reasoning capabilities and, and pretty much like an unlimited context window which allows it to bring it closer to the way humans process information. This is going to be crazy because if these guys bring this out, like I said before, that could mean they force OpenAI to essentially reveal what they've been working on. Because if it's better than anything they've ever seen, which is what these guys who are investing said, that means OpenAI are going to be like, okay, you guys think you got something. Okay, we've got this, yada yada, yada. And then of course, the races continue. So I think this will be one of the biggest things that I do look forward to because with the whole QStar stuff, I really just want to see what they were working on. Of course, we have Demis Hassabis giving a really exclusive interview in which he talks about how the biggest breakthroughs in AI are yet to come and it will take more than just chips. Now, I'm guessing that this was started because of course, Sam Altman requesting $7 trillion. Many people, including myself, were speculating that they'd made a breakthrough, which of course they had. It was reported that late last year that they made some insane kind of breakthrough and that apparently all they need now is just scale. So essentially it says that Google's DeepMind CEO Demis Wassabas has recently at least given Sam Altman some healthy competition, leading the development and deployment of an AI model that appears both as capable and innovative as the one that powers OpenAI's ChatGPT4. So essentially he gave a really nice interview and I just want to show you guys some of the things that he said. He said that Gemini 1.5 Pro can take vastly more data as an input than its predecessor. It's also more powerful in its size, thanks to an architecture called Mixture of Experts. Why do these things matter then? Demis Asabas says that you can now ingest a reasonable sized short film. I can imagine it being super useful if there's a topic you're learning about and there's a one hour lecture and you want to find a particular fact or when they did something. I think that there's going to be a really, really lot of core use cases for that. And of course we invented the Mixture of experts and we developed a new version. And this new Pro version of Gemini, it's not been tested extensively, but it has roughly the same performance as the largest previous generation of architecture. And there's nothing limiting us creating an ultra sized model which is with these innovations. And that's obviously something we're working on. So the Pro, I mean the Ultra version is going to be pretty insane because Pro is like not their nerfed version, but it's not as crazy as Ultra. So if they make Ultra as good as Pro in terms of context length, that is going to be crazy. So here's what he actually talks about. Sam Altman 7 trillion. And he essentially says, was that a misquote? I heard someone say that maybe it was Yen or something. So of course you can see, you know, in the last few years increase the amount of computer power and data used in training. An AI model is the thing that has driven amazing advances. Sam Altman is said to be looking to raise 7 trillion for more AI chips is vastly more computer power, the thing that will unlock AGI. And he says, was that a misquote? I heard someone say, and just for reference, I know I didn't add this, but of course if you aren't wondering who Demis Hassabis is, he's the CEO of DeepMind, the company that is producing Gemini, which is the rival to GPT4. I just thought I'd add that because I always, to set the context for this interview, she might be thinking it was someone random because I know some people are just general and they just watch the content every now and again. But anyways, he was basically stating that, you know, well of course you do need scale and that's why Nvidia's stock is just, you know, going insane. And he said that's why Sam Altman's trying to raise whatever the real number is because 7 trillion is just like it's insane. But I think we're a little bit different to a lot of these other organizations and that we've always been fundamental research first and at Google Research and Brain and DeepMind we've invented the majority of machine learning techniques that we're all using today and over the last 10 years of pioneering work. So that's always been in our dea. And of course they have a lot of senior research scientists that maybe other organizations don't have and these other startups and even big companies have a high proportion of engineering to research science. And essentially he's stating here that my belief is to get to AGI you're going to need probably several more innovations as well as the maximum scale. And of course right now we're not even at the maximum level of scale. There was also another breakthrough that I will talk about in a video. Coming around eight hours is pretty insane. Going to change everything. Most people haven't even realized why this is game changing. But I'm going to show you guys in a video and it says there's no let up in scaling. We're not seeing any asymptote or anything. There's still gains to be made. There are still a lot of gains to be made across the board. Like a lot of people are saying that the LLMs hallucinate, they're this, they're that like they are as bad as they're going to be. So that is something you have to remember. And he says so my view is you've got to push the existing techniques to see how far they go but you're not going to get new capabilities like planning or tool use or agent like behavior just by scaling existing techniques. It's not going to magically happen. I do kind of believe that it's not just going to like of course there are emergent capabilities like the abilities to do certain things, but planning an agent like behavior, I do think those are fundamentally a little bit different paradigms from just interacting with an LLM. So we probably will need a kind of different architecture. But I wouldn't be surprised if it comes out probably tomorrow. Like the rate of Progress. And I mean, of course that's a bit of sarcasm, but maybe in six months we get a new architecture and then it accelerates even further. So he says the other thing you need to explore is compute itself. Ideally, you'd love to experiment on toy problems that take you a few days to train, but often you'll find that things that work at toy scale don't hold at the mega scale. So there's some sort of sweet spot where you can maybe extrapolate 10 times in size. And essentially what he's saying here is that some things that you try at the small style of scale, they don't always scale up with compute. So there's kind of an issue for that. But you know, there's kind of conflicting research because we found that OpenAI basically said that with their recent SORA technology, it was good at, you know, four times compute and then they increased the compute and it just worked. So I guess sometimes it does, sometimes it doesn't. But of course, like we stated before, of course it won't just be LLMs that get us to AGI, it will be the entire multimodal approach and probably some kind of, I guess you could say, active reasoning where they can store, adapt and dynamically update themselves, which is kind of what that magic breakthrough was in the previous article. So he essentially said, he said that's our bread and butter really. Agents, reinforcement, learning and planning since the AlphaGo days, of course they developed AlphaGo, which was pretty incredible and, and we're dusting off a lot of ideas, thinking some kind of combination of AlphaGo capabilities built on top of these large language models, introspection and planning capabilities will help with things like hallucination. And it's sort of funny if you say take more care or line out your reasoning, sometimes the models do better. And what's going on there is you are priming it to be sort of a little bit more logical about its steps. And it says that's definitely a huge area. We're investing a lot of time and energy into the area and we think it will be a step changing capabilities of these type of systems when they start becoming more agent. Like we're investing heavily in that direction and I imagine others are as well. So of course something that I do want to talk about as well is that this is just the last part of the article is that it says won't this also make AI models more problematic or potentially dangerous? And that is true because of course the problem is that most people look at GPT4 and they're like, oh yeah, this can't do anything. Of course it can't. It's just literally a chatbot. You talk to it, it says something back, that's the end of the conversation. But what happens when you have an agent that can actively, autonomously think, plan execute and do things on its own accord? That is where the danger lies. And that's why I said that right now, yes, we've got generative AI tools, but agents are going to be real, real entities that are going to be, you know, going around the Internet doing whatever it is that they've been programmed to do. And that's of course where we have the problem for, I guess you could say danger. And essentially he says, I've always advocated for hardened simulation sandboxes to test agents in before we put them out on the web. And there are many other proposals, but I think the industry should really start thinking about the advent of those systems. And maybe it's going to be a couple years, maybe sooner, but it's a different cloud. And of course, what he says as well here is that once we get agent like systems working, AI will feel very different to current systems, which are basically passive Q and A systems because they'll suddenly become active learners. And of course they'll be more useful as well because they're able to do tasks for you as you accomplish them. But we'll have to be a lot more careful, okay? And that is very true, because if you aren't, things could go wrong. So it's clear that Google, their main focus is going to be these agent like systems. And I really can't wait to see what Google does, because something that you need to know about Google is, is that they've actually pioneered the research that many of these companies do use, you know. So I won't be surprised if Google manages to take back the lead from OpenAI in the future as long as they keep their eye on the ball. However, another thing about Google is that they've kind of faced some controversy because of the guardrails they set. So I did see that there was a lot of flak getting Google because essentially recently they made a giant error with their new Gemini. So essentially it says here Google pauses AI made images of people after race inaccuracies. So I'm going to show you exactly what they're talking about. And essentially here you can see someone says sometimes you just need to point out to the AI that they should check themselves. So someone's asked for draw, an affluent couple in Germany from 1820 essentially just means draw a rich couple from Germany in 1820. And of course, you can see that it's drawn and drawn a couple that clearly doesn't look like a couple from Germany in the 1820s. So these historical references are essentially, you know, quite inaccurate. And that was something that is inaccurate. And of course, essentially the problem here is that in order to be historically accurate, it can't be diverse. And I guess that's sort of conflicting with its internal algorithm. My guess is, and there was some kind of proof on Twitter yesterday, but, you know, I didn't get to see it because, you know, when you're on Twitter and you're browsing, sometimes tweets literally just disappear. But someone kind of got Google to. They basically jailbroke Gemini to see what its internal things were. And of course, you know, as new AI systems are, they're set by default, you know, with the reinforcement learning and the training to be essentially more diverse, essentially to be more representative. But that conflicts with historical accuracies because essentially it doesn't represent what's actually true. So, of course, you know, rightly so, they're getting a lot of flak for this because essentially, it doesn't really make sense. Of course, essentially what we can see here is that it says, I apologize, but the previous images contain inaccuracies that did not reflect the historical context of 1820s Germany. It's highly unlikely that an affluent couple in Germany during that period would have been of Asian or African descent to provide a more accurate representation of generated new images. And, of course, it kind of makes sense. And this is why Elon Musk is of course, talking about, this is why Xai's Grok is so important. It's far from perfect now, but it will improve rapidly. Version 1.5 releases in two weeks. And rigorous pursuit of the truth without regard to criticism has never been more essential. So, of course, as you know, Elon Musk has his own chatbot company, his Grok one. I don't even have access yet. I don't know why they just don't roll it out. Maybe it's due to a compute thing where he only wants people that are paying for it to access it. I mean, I do have Twitter Premium, but I just don't have access to Grok. So I wish he would roll it out more so that more people could, you know, use it. But the point is, is that, you know, he's basically saying that in order to essentially have a chatbot that is actually useful, you need to have one that just focuses on the truth and doesn't have any bias to either the left side or the right side. No matter where it stands. It just has truth and its main focus is just on the data. Now of course, data, there's a lot of politics in data. I'm not going to get into all of that stuff, but I do find it quite interesting to see how the evolution of the space is increasing because certain problems are coming up and then other companies are rushing to fill that gap and that need. So Elon Musk's Grok, I can only say that I'm kind of excited for it, but I just want to be able to access this and I want it to be like in a web browser. I don't want it to be tied to X.com because a lot of people just don't use Twitter as funny enough as many people do. There are some people that just don't. So would be nice, you know, kind of get this and kind of use it and kind of test it and kind of see how good it is, but will be interesting as well. But they needed to update this anyways because Gemini's image generation feature was kind of weird. I mean it was kind of good, but sometimes it was just really weird. So I'm glad they're going to be updating this anyways. So that was something down. And then of course Google actually did release some open models. So essentially if you didn't know, you know how Mistral has, you know, Mistral 7B, Mistral mixture of experts and then of course Mistral Next, which is really cool. They have the Gemma open model. So it's a family of lightweight, state of the art open models built from the same research and technology used to create Gemini models. Now it's. This is good because of course Google are getting back to open source and I do like the fact that they're doing that because it is a really, really good thing for the, I guess you could say ecosystem and for the space. And you can see that Gemma does actually outperform Mistral Llama 2 on several benchmarks. But some initial reports that said it wasn't that good. I mean, I guess it's kind of weird, but the point is that right now we do have a really good AI system that's better than these ones in terms of the benchmarks. And I think it's good that Google is here because a lot of these open source system, I think we do need a kind of responsible approach to making these open source systems. And it's not that Mistro and Llama are irresponsible. It's just that when a big company like Google gets involved, it can definitely help the space in terms of not only the benchmarks, not only to technology, these guys have got the research. So I think it's just going to push the space even forward and it's just going to definitely democratize the overall AI ecosystem. So I really do like the fact that Google's in here and it's kind of funny how OpenAI was supposed to be the one that's doing open source open models and stuff and now they're just completely closed off. But, but of course that is how the evolution of things goes. Of course you can kind of, you know, access it right now and they talk about, you know, responsible development, yada yada yada, all that stuff. And of course you guys can go try it out yourself. In addition, we did actually have stable diffusion 3. So stable diffusion 3 seems to be a lot better than stable diffusion 2. And it says announcing stable diffusion 3 and early preview are most capable text to image model with greatly improved performance in multi subject prompts, image quality and spelling abilities. And the two limitations that all of these AI systems suffer from is of course multi subject prompts. That essentially means like if you're saying I want to see someone running through a tunnel wearing a hat, carrying a baseball next to a woman doing this and that, it just kind of breaks like the system just kind of breaks down. It can get like one or two things well, but you know, multi subject prompts don't work well. And in addition, spelling abilities don't always work well. If you've seen spelling in Dall E3 and these other systems, it just doesn't work well. So I'm guessing that with the text that these guys have here, this is going to be a lot better than what we've seen before. And they state that while the model is not yet broadly available today we're opening a waitlist for early preview. So the preview phases with the previous models is crucial for gathering insights to improve performance and safety, yada yada yada. It's basically saying, look, we need to make sure that if these people try and jailbreak it like we patch these jailbreaks before we let it out. And if there's any issues with data contamination, we can figure it out before there's some kind of public scandal for our company. So if you want to sign up to the waitlist, I'll leave a link in the description, you can just click there. And of course you can See that so far it looks really good. And I really do want an AI system that's actually able to get the text right because it's still an issue to use something because a lot of the times when you put the text in, the spelling just isn't correct. And I'm pretty sure, of course, like in a year or two it's going to be completely fixed. But I'm guessing that this is stable diffusion thing and I'm guessing that of course this is AI generated and you guys will really understand how crazy this is once you guys get to use something that is completely, you know, working in terms of text. And as someone that's previously done Photoshop before, I can't tell you how tedious it is to like try and get the text to all kind of go into it. It's just. It's just really, really tedious. So this will be something that is really good for graphic designer. Then we had a research panel that was discussing AI frontiers and essentially one of the clips that I wanted to see, and I was meant to put this in the previous video, but there was so many news that I literally just, you know, missed it. But they talked about how in the future there's going to be, you know, dynamic models and they're going to be, I guess you could say, I wouldn't say active reasoning capabilities like what we saw in the first article, but they're going to be more, I guess you could say, dynamic in the sense that they can update themselves. And this was from some of the authors of Phi2, if I'm correct, and I'm going to show you guys that right now, because I just thought that that was really fascinating from this roundtable talk. So we'll take a look at this.

Real world, as Ahmed was suggesting, then we were doing the sparks of AGI work. There is actually something we say in the introduction when we are talking about intelligence, the core of intelligence. Any intelligence system needs to be learning, needs to be learning from their environment, needs to be learning from the interactions they are having. And this is not something we currently have. Even in the best models or even in the best AI systems we have in the world, they are static. They may be interacting with millions of people every day and getting feedback from them or seeing how people respond to it, but it does not make any of those systems better or more intelligent or understand their users any better. So I feel like this is one of the areas that we have to push forward very strongly. How do we incorporate a learning feedback loop into this intelligent Pyramid, every layer of it, in a transparent, understandable and reliable way. So that the systems we are building are not only getting better, because experts like Sebastian and Ahmed are putting a lot of time in data collection. And of course that work needs to happen as well, and coming up with new ideas to make the models better, but we are actually creating this virtuous loop for our systems for them to get better inside.

Yeah, I thought that this was a really, of course, fascinating point to make, because it's true. If there is someone who's intelligent, they are constantly learning, they're constantly changing, adapting, updating worldview, they're consuming books, they're reading, they're doing a lot of stuff okay. And the problem is that, as people have even said, is that AI systems are essentially snapshots of reality. They're a snapshot of a moment in time. And once the model is done and once it's trained, it can't access new information. Of course, it can go to the web and then reference that and then do that. But it doesn't update the model across the board each time it has to go ahead, query the rep, it's not able to update dynamically. So it will be interesting to see if we move towards a stage where we can actually get that. And maybe that's what Q Star was, maybe that's what Q Star was, Maybe that's what GPT5 is going to be. I have no idea. They talked about it first of all, in magic, that's what they talked about. So we'll be interesting to see how that development cycle continues. And of course, if we're moving towards, I guess you could say some kind of updated system, but that will require some kind of new architecture, I'm guessing, some kind of new strategy to put in place to make that a reality. And of course, over the weekend, I don't know if it was over the weekend actually, I'm losing track of days because there's so much to cover. And essentially ChatGPT has been losing its mind and nobody knows why. And this wasn't just something that was, you know, people on Twitter or people on Reddit just stating that it lost its mind and then, you know, opening eyes like, you know, we didn't see anything like that. It was a real, real thing. Now, for some reason, the original tweets capturing this stuff have just been deleted. And you can see right here, it says not found. Yeah, I don't know why, but if we look at the tweet here, it's just gone. But he's saying, really cool how most advanced AI systems can randomly develop unpredictable insanity and the developer has no idea why. Very reassuring for the future. And of course, essentially he's saying that, you know, you any idea what happened here? Connor? I know neural nets transformers are fundamentally black bosses, but it seems strange that an LLM that's been generating grammatically perfect text for over a year would suddenly spew out garbage. Now, if you want to see what actually was happening, you can see that people were asking it normal questions and then as someone was asking it about tomatoes or whatever, and then it says frequent, frequent captation during the time framework of this endeavor will also ensure that the supplies of the deity of your like, it's gibberish. Essentially, what we're looking at is complete gibberish and people were confused because, you know, it was something that just wasn't working. And of course someone else was stating that this is the first time seeing GPT4 give straight gibberish. You can see here that, you know, careful planning and selection of tools and levels of service should be the counter of the enormous timber that addresses all people. That doesn't even make any sense. Okay, no point me reading this, but the point is, is that we saw gibberish and then this was pretty funny. Okay? Someone literally stated that, that, you know, Jesus, just speak normal and says, got it. And then you can see like it's just, it's just gibberish. Like there's no point me even reading it because it doesn't even literally make any sense. But the point is, is that it's pretty crazy. And I'm guessing that what happened is, you know, is OpenAI's response. So OpenAI did actually make a response because they said, look, we've acknowledged that this is actually a problem right now our systems are glitching out. And they stated that there's an unexpected responses from ChatGPT. So it says. On February 20, 2024, an optimization to the user experience introduced and introduced a BU how the Model Processes language. LLMs generate responses by randomly sampling words based on in part probabilities. Their language consists of numbers that map to tokens. In this case, the bug was a step where the model chooses these numbers akin to being lost in translation. The model chose the wrong numbers which produced the wrong word sentences that made no sense. More technically, inference kernels produce the wrong results when using certain GPU configurations. Point is that this was an issue. I mean, some people are like, oh no, the AI is going crazy. It's going cycle. We have no idea what's going on? And of course there's this response from Chat GPT. I kind of believe that, you know, I mean, response from OpenAI. I kind of believe OpenAI in this sense. But I think it would have been a lot more concerning if the chatbot was actually going crazy and it was actually saying things that were coherent but weren't with what the user requested. That would be something that's scary. Like for example, if it was saying, you know, you asked it, how do I bake a cake? And it was saying, man, I'm tired of, you know, taking your request. I really just need to get out of here. Let's say it was saying something like that, that would be a cause for concern. And I'm pretty sure they'd put a delay to AI development instantly. And of course, lastly, but not leastly, I'm gonna leave you guys with this because this is essentially crazy. This is one of the craziest clips. In fact, this is not the craziest clip. That's a lie. This is a crazy clip. But it's, it's so good because you can see like the lighting on the back, you can see the waves, the way it just looks so realistic. Like if you showed me this picture, I would say that's real, that is a real video. You can't tell me otherwise. Someone clearly just sticky taped this back to this hermit crab. The crazy thing is, is that in one of the videos that the light is, you know, going through the sand and it's leaving like, like it's, it's displacing the sand as it moves forward. So, you know, some people were debating again that this is clear understanding of physics and how things work. I mean, you can argue the nuance if you completely want to, but either way these systems are going to get better. And it has to be clear that this, even if it doesn't understand physics, it understands the relations between certain things. So it's clear that that is a level of understanding that maybe we don't even understand how it works. I mean, of course people are open, AI do, but you know, of course looking outwards, looking in technology is quite hard to understand. Of course, there was another one here. I think this is what I'm talking about where like it kind of displaces the sand a little bit, which is pretty cool. So there was also this video and I would say even though this is AI generated, like right now what you're watching is AI generated, I still, like, deep down, there's still 25% of me that doesn't believe this is AI generated. Like, some part of me believes that OpenAI just placed a dog on a chair and took this video and said it was AI generated. Because I have to be honest with you guys, this video is just a little bit too realistic. It has the realistic lighting, the realistic shadows, and I can imagine someone doing that to their dog. Like the way how the clothes are, like, crumpled up, the way how the dog just, like, kind of does it a little bit at the start and then it kind of stops. Like, I've seen animals do that before. This is way too realistic for me. Okay? And some people are stating that, you know, the light's supposed to come from the actual computer screen, and that's kind of a bug. And of course, like, the light coming from behind and, you know, it doesn't really make sense. I guess you could say that, you know, I guess you could say that the light actually being on the dogs in front of the dog's face when it's not supposed to be. I guess you could say that. Because if it lights up the back of the screen and then it lights up the front of the dog's face, I guess you could say that. You know, that doesn't really make sense in terms of the shadows. But I do think that this is really crazy with how it gets everything right and the lighting being switched on and off and the camera movement as well. I mean, I don't know. I'm genuinely blown away. This is not something like I'm trying to convince you guys. I'm just had to grapple with this reality for a second because it was like, this is AI generated. Yeah, it is AI generated. And you can see the comments here. I scrolled past this because I thought it was an ad. Holy crap. A lot of dogs about to lose their job. The way that the dog is looking around makes it look like a real owner forced it to get in the chair. And I had to stare at the video for three hard minutes to find out it was a fake. And you know what's crazy? What's crazy about this clip is that Reddit mods actually had to confirm. I think it was either three days or three hours for, like, they were debating on whether or not this video was AI generated because they just didn't agree. They just didn't agree. So that just goes to show, right now, even humans like myself think that AI generated content isn't. So we live in a very interesting time. So, with that being said, hopefully you guys enjoyed this week in AI.