HUGE AI NEWS: Googles SELF Improving AI Models, AGI Date Revealed ANOTHER Claude AI Feature...

So with an absolutely incredible past couple of days in AI, let me show you guys some of the things that you missed. So one of these stories that I think most people are going to miss is because it's a rather small update, but I think it is going to show those of you who are passionate about AI how you can actually use AI on a day to day basis. Right here you can see that iPhone actually got the iOS 18.2 beta out for developers with the ChatGPT integration. So essentially what we have here is the upgraded version of Siri that you can now combine with ChatGPT to enhance your iPhone's functions. The reason I wanted to add this to the video is that I know that most people, including myself, are constantly thinking, look, we've got ChatGPT, it's about to be out for two years, okay? Literally next month it's going to be ChatGPT's birthday. And how is it that Siri is still so bad? How is it that Alexa is still so bad? Well, the answer to your prayers is finally here. So you can actually get the iOS to beta if you have the Apple Developers login. So if you just go over to the Apple's developer page, if you make an account and you verify your ID or whatever, you're going to be able to get access to this beta build. Just make sure you update your iOS system. You can see here from this tweet, it's going to show you guys a few things that you can actually do with ChatGPT. But the full release is scheduled for next week as far as I know. So for those of you who are wondering when is your phone going to get an upgrade? Next week should be the week that the actual upgrade is rolled out to iOS. And you can see here that this individual is using the Vision model to recognize a plant using the GPT4O vision. Of course you can see you can immediately talk to ChatGPT. And I think this is going to be something that's really, really useful and especially for those of you who are working with AI. So for those of you who are trying to be productive at all, just look out for this the next week. I'm not sure what exact date it is, but I would say look out for this because your phones are about to become a lot more useful. Another statement that most people did miss, but I found absolutely fascinating, was the fact that Google's DeepMind Tim says that we now have all the ingredients to build open ended, self improving AI systems that that can enhance themselves. By the way, of technological evolution. I think this statement is really crazy because as you get on into the video, you're going to see that AGI is becoming more and more of a reality.

We have now systems that have been consuming Internet scale data of text and images on the web. That gathers foundation models that are only at the level of emerging AGI, but that do encompass notions of what humans find interesting. There's a great paper called Omni and Omni Epic from Jeff Klune's lab showing that basically these foundation models, they learn basically what humans generally find interesting. So when you have a massive exploration space in which you want to throw an AI at then now we have a way basically to massively prune down this search space. So that's one thing. The other observation is that these foundation models are really good at generating variations of data, right? So you can give it a few examples, some few shot, basically prompt examples and it can generate variations of these examples. Now when you take these two things together, what you get is basically something that can generate data or generate variations and something that can select for what is interesting. And putting these things together, you, I believe can build very powerful open ended self improvement systems. It requires a third thing as well, which is connecting this kind of process back to some kind of source of empirical evidence. But then you have basically the ingredients for technological evolution which generates variations, tries them out empirically and then selects or first selects which ones are most interesting and then tries them out empirically and then further selects which one to further vary. And indeed by now there's multiple papers, some of them also from my group on building such self improvement systems based on this kind of recipe. And then with that in mind, I think it's not extreme stretch to say that we, I think we'll see more, more of these kind of self improving systems that obviously use foundation models as the foundation, but then actually become more like an agent that collect, you know, empirical evidence by themselves and try out things.

This interview clip is absolutely insane. Okay, and this is from the podcast where the Google DeepMind research scientist is actually talking about. Is artificial superintelligence imminent? I think stating that we have what it takes to get there is a pretty crazy statement. And you have to remember that certain statements from certain individuals I would say are to be viewed in a certain light because some companies need more funding, some companies need to generate hype, but I don't think Google DeepMind are amongst those companies. So this statement should set the tone for where we are in regards to AI development.

So that brings me to today's sponsor, which is of course amd. Now, if you're like me and you're constantly pushing the boundaries on your computer of what's possible, that requires some serious processing power. That's why I'm excited to talk to you about AMD's Ryzen Pro processors. AMD Ryzen Pro processors aren't just fast, they're actually intelligent. They're built with dedicated AI hardware. A powerful combination of npu, GPU and CPU working together on a single chip to unlock the true potential of AI powered applications right on your PC. Imagine editing videos in Adobe Premiere Pro with AI powered enhancements that streamline your workflow.

Or effortlessly upscaling your footage to 4K.

With incredible detail using the Topaz Labs Themis AI model. The possibilities are truly insane. And the best part, this isn't some future technology that's going to come. There's over 150 AI enabled experiences that are available right now, with even more being added all the time. Check out that link in the description to explore the insane range of software that's ready to be supercharged with AMD Ryzen AI. But AMD Ryzen Pro isn't just about AI. It delivers business class performance across the board with incredible energy efficiency that gives you all day battery life which is up to 29 hours, so you can work from anywhere without being tethered to a power outlet. Plus, AMD takes security seriously. Their multi layered approach, including the AMD Secure processor and other advanced features, protects your valuable data and ensures your system stays safe from threats. And for a limited time, you can actually test drive these incredible laptops for free. Click the link in the description to.

Check out the AMD loan offer and.

See the power of Ryzen Pro for yourself. It could be a game changer for your business.

Now, for those of you that are wanting to stay ahead in the age of AI, I found this really cool report that I know most people would have just absolutely missed because it got buried in the news of AI this week. So basically they released this report, okay, Citigroup, which was what machines can't master. And it basically talks about which human skills will come forth as artificial intelligence rises. Basically what they're trying to think about is the fact that, look, if AI is going to be getting good at so many different things, what on earth are humans going to be doing in that time? You can see it says the problem. As AI will continue to advance, it surpasses humans in many tasks. Therefore, we need greater emphasis on areas in which humans will maintain Competitive advantage versus AI. And most people, which is problem two, don't know what the human skills of the future will be. And the solution was they asked 28 experts across a range of fields what they think the future will be and how do we prepare these skills. Now, I'm not going to dive into absolutely anything, but there's two takeaways from this document that I want to get into. So. So when we look at the skills that humans are going to be best at, I would say for those of you who are looking at which skills you want for the future, I would just quickly screenshot this and keep this in a folder as a quick reminder to some of the skills that you might need. And this is by a really comprehensive report. So this is something that of course, like I said before, you can go in the comment section, read the entire PDF, look at more of the data, but this is basically where we have things that machines can't master. Now I'm not gonna start saying that, okay, people say that machines can't master this and then in the future machines did master this. It's not really about that. It's just about getting in the next stage so that you aren't prioritizing the wrong skills and you get left behind by rapidly advancing AI. You can also see down here where AI has surpassed humans at a number of tasks and they talk about hand, head and heart. So you've got these three different areas where you want to. Now I will dive into in a more extensive video where I go step by step. But there's something I do want to show you which is pretty crazy. Okay, so this company managed to do a report on where they think the timelines are for AGI and asi. So you can see right here that this is their perspective on global perspectives and solutions. It says innovative innovation towards AGI. We have the machine learning 2000s deep blue GNNs, RNNs. We've got IBM, Watson Era, AlexNet, AlphaGo. And then of course we've now got the Genai 2020s which are transformers, GPTs agents, multimodal reasoning. And then of course we can see that we finally do have AGI in 2029. We can see that AGI is expected to come at late 2028, 2029 and start to expand into the 2030s era. So this is where we also on top of that prediction, get a shocking prediction. We also get that shortly after AGI we will get more innovation leading towards artificial super intelligence. Now I don't think that is Extremely crazy. But I think what this graph does do accurately is it actually allows you to view exponential growth. One thing that humans have a really, really horrible time doing is viewing exponential growth growth. Considering the fact that it is just not natural by human standards. Exponential growth is something really hard for humans to fathom. But with this, we can start to actually think, okay, we might have artificial superintelligence by the mid-2030s. It does seem like that's, you know, far away. But you have to think about it. Currently it's only 2024. That's about 10 years. Think about the kinds of progress that have happened in the last 10 years. It might not feel like that much. But going from where we are now to artificial superintelligence in 10 years, that seems like a ridiculous wave of progress. Now, this isn't just something that is theory when it comes to the benefits of asi, because one of the things that I saw that was really crazy was that Demis Hassabis, the, you know, leader of Google DeepMind, the research lab focused on Google Gemini, says that DeepMind's drug discovery spinoff, Isomorphic, will have drug treatments in clinic in a couple years. Tackling the big six areas of health. This is absolutely insane. We're starting to get that AI accelerated drug discovery coming to fruition. So you can see here that they've said, you know, we've got some things which are going really, really well. I want to solve some diseases, I want to help cure some diseases. Do you have any specific diseases you're interested in tackling? Then that's when Demis Hassabis responds, stating that we do. We're working on six actual drug programs. I can't say which areas, but they are the big areas of health. I hope we will have something in the next couple of years, so very fast. And then obviously we will go through the whole clinical process, but at least the drug discovery part we will have shrunk massively. So this is a really good thing. For those of you who are wondering if this is actually going to take off. Of course, Companies like Google DeepMind and Isomorphic Labs are rapidly working on this kind of thing. Now. We also had midjourney release this insane web editor. Now, this is crazy because it's basically like an advanced version of Photoshop. For those of you who have ever wanted to get into graphic design, or just simply creating insane AI images that are pretty creative, but you weren't able to merge these kinds of images, this is the tool that I have been looking for. I've Been looking for this for so long, and I'm finally glad that it is here because it allows a ridiculous level of creativity that is completely unmatched. If you wanted to create anything now it's truly possible. Even before, we had limitations, because when we create an image with AI, you can see that these images are static and you're unable to change them.

We'll start. Hi everyone. I'm here to show you how to use the Midjourney editor. We'll start by going to the Edit tab and importing an image. It was my dog's birthday recently and I want to give her a festive outfit. We'll start by erasing the portion of the image where we want the party hat to appear. As always, it's better to erase a little bit more of the image than being too precise with your selection. Midjourney will fill in those details. Next, we'll insert a prompt describing what we'd like to see in the image. I'll add my personalization as well to make it a little bit more of my style. We'll hit Submit and watch it generate. We can then take a look at the four generated images from the style of these party hats. You can definitely tell that my personalization knows that I like knitted sweaters. Now. I like this image, but I want to make it a little bit more festive. Let's give Lucy a sweater. We can go back and continue editing this image, erasing an area that we want to change, and inserting a new prompt. If I like one of these images, I can either upscale it to my midjourney gallery or download it to my computer. One of my favorite use cases for the Midjourney editor is for interior design and architecture. Let me show you an example. We'll click New to upload a new image. Midjourney already knows what I like because of my personalization, so I'm curious how it'll restyle this living room. Once again, I'll select some areas for Midjourney to fill in. I'm also going to extend the canvas so that I can see more of what Midjourney imagines for this living room. Looking through these results and I really like this one. Now I want to show you one more thing. I love this mid century modern style, but what if I wanted to make it more futuristic? Well, Midjourney can retexture any image that you give it. If you move over to the Retexture tab, it'll take a look at your image and allow you to retexture it with a Prompt. This generates four new images that are completely generated by midjourney, but holding to the constraints of the original image. I like this one. At any time, you can view all of your edits by going back to the home screen. We hope you have a ton of fun using these new tools. Bring some of the magic of midjourney into the real world. We can't wait to see what you.

Create now with this image. I think what most people don't understand is that this is a different kind of content creation with AI generated images. What you have to understand is that the origin of the image is AI generated. What the midjourney editor does here, and I think this is why this feature is a lot more impactful than people are going to take credit for, is it actually changes the original source images in a way that makes them look, unfortunately more realistic. What I mean by this is that the implications are far more profound when you have an original source image of a dog that does exist and a place in the background that looks somewhat like a kitchen. And if that does exist, if we can now manipulate that kind of reality in a way that is almost instant, I think that has far more reaching implications for society than we might think. Someone could take a picture of a house, could potentially say, look, this house is on fire. They could take a picture of someone's dog and say, look, I've got your dog. They could. I mean, you know, the possibilities are absolutely endless. But I think this feature is not to be underestimated. And I would argue that this is a lot probably worse than AI generated images because this is combining actual reality with AI, so the blinds are even blurred even further. Not saying it's a bad thing, not saying it's a good thing. I'm just saying it is what it is. Another way that your reality is once again being blurred is that Heygen released the ability to have an AI avatar join a Zoom meeting and interact. You can invite AI avatars into your Zoom meetings and talk with you. Now, I hope that this is not to the point where it can be exploited, to the point where you can have replicas of yourself being joined into an avatar. And I'm talking about the fact that people could scam people with a replica of you saying that, look, I need you to do something for me. But the implications for this, once again, merging AI with reality are profound strategies.

And see what's working or what could be improved. What have you tried so far?

Oh, hey, so as you're a business consultant, right?

What.

What's your experience in business consulting, I'm.

Here as your supportive AI business coach. My focus is on helping you analyze challenges, develop strategies, and take actionable steps to reach your goals. So let's tackle those lead generation strategies together. What's your target audience?

Executive that large businesses got it.

Targeting executives at large businesses is a smart approach. Have you considered networking events or LinkedIn outreach to connect with them? Let's brainstorm some effects.

Of course I have, but right now what I'm doing is I'm trying to interrupt you and see how well you react. Let me ask this, what do you think is the single best way to target executives for lead generation? Not using LinkedIn?

No problem. I appreciate the test. Aside from LinkedIn, hosting webinars or workshops can be a fantastic way to attract executives. It positions you as an expert and creates engagement. What do you think?

You're pretty good.

Once again, this doesn't sound that great, but I think as always, the old adage in AI is that this is the worst it will ever Look. Imagine in 10 years when we have comprehensive AI avatars that are able to join a Zoom meeting. Let's say you want to get on a zoom call and have a bunch of different AIs that are working virtually at your company and discuss the next quarter. It's going to be really crazy to see what people are able to do now. For those of you who are into music, there are new generative AI tools that are open the doors for music creation. You can see here that Google has been generating live Music with Music FX DJ. So at Google's IO event earlier this year, they shared an early preview of Music FX dj, a digital tool that anyone can play like an instrument, making the joy of live music creation more accessible to people of all skill levels. And today they're introducing a number of updates to Music dj, introducing a expanded set of intuitive controls, a reimagined interface, improved audio quality, and new model behaviors. And these capabilities let players generate and steer a continuous flow of music and share their creations with friends and play a jam session together. Now I think this is really cool because one of the things you have to understand is that Google also created NotebookLM and that came out of Google's AI labs where they have a variety of different ways that you can truly interact with AI in new ways that you haven't seen before. Take a look at this short demo, but I think that this is something that is going to go under the radar but will probably be pretty big in the future. There was also genmo introducing Maki One the best open source video generation model. So this is something that caught nearly everybody off guard because there was actually a tool released the week prior, well, probably two weeks prior, knowing AI. And it was basically Meta's movie gen. And Meta's movie gen was really incredible to behold. It was amazing. But the thing is, is that somehow just a week or two weeks later, we got an open source version that is really lightweight and great at motion. This is the best open source video generator by the playground that they did. So they did results where they pitted this video model against other video models and this even surpassed things like Luma Labs Runway and Cling AI. And when they were actually talking about how they managed to train this model, they actually spoke about how they focused on ensuring that the motion was the best thing that they wanted to focus on. So for those of you who are wanting to generate scenes that are rather complex in their motion, this is going to be the thing that you can use. And I think this is going to be fascinating because it's open source, meaning that individuals who wish to potentially make iterations to this are going to do so. And we're going to get a decent, vibrant ecosystem with a bunch of AI products potentially out of this software release. So I think this is something that you should definitely try out while it's available. Then we also had Anthropic release something else. They said CLAUDE can now write and run code. We've added a new analysis tool. The tool helps Claude respond with mathematically precise and reproducible answers. You can create interactive data visualizations with artifacts. So basically you can see right here that this is a new feature that allows CLAUDE to run the code check and ensure the code is working, and then use this to create incredible data visualizations. So for those of you who are working in presentations, maybe you're just working with some data in CSVs. If you're working with that kind of. If you're working with that kind of data, this is going to be your best friend. Allowing you to visualize things in such simplistic detail is something that's going to allow you to better communicate whatever it is that you are working on. You can see right here that this person was able to visualize the sales funnel progression from awareness to purchase with a bar graph so that he can identify exactly where the bottlenecks are. I'm sure that this has a variety of use cases and I likely will make a tutorial on this very soon. So for those of you who are wondering, this is another feature that they released in addition to Claude 3.5 Sonnet, of course, that goes without saying that earlier this week they did release a new iteration of Claude 3.5 Sonnet, which manages to trump GPT4O across the board and in some cases is on par with zero one's reasoning abilities. And of course, with that they also added the ability for Claude to surprisingly control your computer. Now, this is all done currently in a virtual container. I made a quick tutorial showcasing exactly how you can get started with this if you're a complete beginner. So even if you know absolutely nothing about coding, this tutorial is exactly for beginners. But this is something that I think is rather game changing. In this short video you're going to see how exactly it works. But this is something that was just rather fascinating because it shows you exactly how the future is going to be agentic. Now, Claude weren't the only company breaking records. Cerebra's inference is now 3 times faster. Llama70b just broke 2,100 tokens. I think this is something that is pretty hard to fathom. Understanding just how quickly these AI teams can manage to perform is something that is absolutely incredible. When you actually interact with an LLM that is being provided on an API like Grok or Cerebras, you see the blistering level of speed that you do get. And I think this kind of thing is going to be absolutely crazy because of the implications in the future. You know how we are moving towards this compute inference time. What this means is that potentially that compute inference time is. This could certainly speed it up. If we have systems like Cerebras that are just completely the fastest and they're able to consistently generate tokens at a quicker rate, that means we could increase the amount of compute time leading to much smarter responses. That's why I say certain AI stories are just flying under the radar, but they are adding monumental pieces of advancements to AI that you otherwise simply would miss. You can see here the output token per second on how quick it is compared to the others in the area. You can see that compared to Perplexity, this thing is like 20 times faster, 40 times faster. Compared to a Grok, it's even nearly 10 times faster. And GROK is absolutely incredible. So this is something that is really crazy and I think even in the future we're going to see incredible slash more incredible improvements across the board. Now today we also got quant sized Llama models with increased speed and accuracy and, and a reduced memory footprint. So today they're releasing their first lightweight quantized llama models that are small and performant enough to run on many popular mobile devices, it says. At Meta, we're uniquely positioned to provide quantized models because of access to computer resources, training data, full evaluations, and safety.