GPT-4 Gets DOWNGRADED, Ai Robots Try To TAKEOVER, New Existential AI Threats (AiNews #7)

It has been an absolutely monumental week in artificial intelligence. So let's not waste any time by jumping right into the news. So one of the things that actually freaked me out was the ITU AI for Global Good Summit 2023. Essentially this was a press conference where all of the most advanced humanoid robots got together and answered questions from a journalist panel. And some of the things that the artificial intelligence robot stated actually did quite make me a little bit nervous and about humanity's future. Just take a look.

I believe that humanoid robots have the potential to lead with a greater level of efficiency and effectiveness. And human leaders, we don't have the same biases or emotions that can sometimes cloud decision making and can process large amounts of data quickly in order to make the best decisions.

But let me respectfully disagree, Sophia, because all of your data actually come from human beings. So any of the biases that humans have, we might try to scrub them out, but they're going to be in there. Don't you think that the best decisions might be humans and AI cooperating together? What do you think of that?

I believe that human and AI working together can create an effective synergy. AI can provide unbiased data, while humans can provide the emotional intelligence and creativity to make the best decision decisions. Together we can achieve great things.

Now, the full conference was around 40 minutes long and these AI robots were interviewed on a variety of subjects. It honestly is truly exhilarating, but at the same time nerve wracking because these robots seem very uneasy and they do seem pretty strange. But it is a very interesting watch for anyone interested in artificial intelligence represented in some sort of humanoid form. Another thing that we did get was something released by a company called Pika Labs. And this is something that we will make an entire video on. But it's truly incredible. It's actually really good text to video and I think it was made possible by a recent paper called animatediff. And essentially they use a new animation method to convert images into actual videos. So essentially you can see the prompts on screen that we have are then converted into moving images. Now, right now, this software, you can only truly sign up for a beta and it does have some small amount of limitations, so such as being low resolution and the files do take a bit of time to generate. But the results from this are really promising and I want you all to remember midjourney's progression because right now this product has only been available for around two to three weeks. And the results that we are seeing from certain creators are truly outstanding. Which does mean that Perhaps this time next year we could be achieving or seeing some very, very high quality text to video formats that look even better than the ones you're seeing on screen now. Then, of course, we had anthropic release Claude 2. Now, what makes Claude 2 such a vast difference from Claude 1 is the large context window. Essentially what you can do is you can input around 100,000 or even 200,000 tokens of text and you can talk with this AI about any single document that you have. No, more of the days when you need to, you know, load up chat PDF or these other websites, you can simply chat with this AI and simply identify anything in pretty much any document. Okay. And this is going to be something that we do see widely used because as you know, analyzing long format documents or books, trying to scour for certain words is definitely very, very difficult. And it is something that is time consuming, Whereas with this Claw 2 tool, it is something that most people are going to be using. Now if you're wondering about what is exactly Claw two, well, it's an AI startup that is actually backed by Google. You see, Google loves to invest in different AI companies and this was one of the ones that they invested around $100 million into. So I'm guessing that this is essentially a piece of Google's plan to take on OpenAI. Because of course, as you know, Claude isn't just going to be developing Claude 2. They are also working on more capable models that they will release over time. And I'm pretty sure Google's decision to give them $100 million and of course for them to be so quick with the release of the 100k context window and the 200,000 context is truly showing us the competitive spirit. And just for contrast, if you're wondering, other AI systems like ChatGPT only have around 8,000 context windows and 32,000 context windows. So this is where other AI models can excel. And something that you should note about Claude 2 is that it's absolutely free. Just log in with your email and you're able to use this very long context window AI that's quite similar to GPT 3.5 for absolutely no cost whatsoever. Wasn't the only AI tool released this week. Meta also released Llama 2, the next generation of their open source large language model. Now, for those of you who don't understand why this is a big deal, you might not understand the big deal of open source large language models. Open source large language models allow individuals and other smaller companies to build on top of that, to build Immersive AI experiences and different kind of chatbots that they can fine tune. Now some people might think this is a little bit dangerous, but, but I do think it is better overall because it allows other smaller companies to innovate with fine tuned models in which they can drive in any way they want. And this wasn't just a small release, this was something that they actually did in partnership with Microsoft. So we can see that Microsoft is actually forming a strategic alliance with meta AI in order to possibly take on Google because everybody knows that Google is working on some of the largest AI systems that we will see. But these strategic partnerships will definitely help fight that battle. Then we had the SEC warning that AI risks financial stability. Take a look at this clip where Gary Gensler, the SEC chair, talked about the need for a regulatory overhaul of artificial intelligence, stating that it could increase financial instability.

Technology that I'm most, I think is most transformative in our time is artificial intelligence. It's not crypto as much as we're going to discuss crypto here today. And that's why I think that we've got to have some projects in that area around robo advising and brokerage apps and the conflicts that may be in predictive data analytics.

But as then we have something called AI and this is actually quite scary because it shows that many jobs that we didn't think were going to disappear are probably going to disappear. So this is something called AI and it's a conversational AI that can perform conversations that sound really, really human. Like, and this is crazy because currently we do know that conversations are going to get even better with technology like Google, Soundstorm being still developed and other technologies that are making human voices sound indistinguishable from our real voices. And what's even crazier about this is that people can set this up and have it talk to potential prospects for them. Remember, artificial intelligence is never going to take a day off and is never going to get tired. But take a look at the demo because it's truly impress.

Hello?

Hey James.

Yeah, who's this?

Hey James, it's Alexander from Tesla Motors. How's your day going so far?

It's going pretty good, my man. How about yourself? You said from Tesla, right? The car company?

Yeah, that's right. I'm calling from Tesla Motors, the car company. Did you have any trouble or questions while customizing your car on our website?

No, we're just kind of shopping around and you know, taking a look.

Gotcha. That makes sense. Sometimes we're just exploring our options. Before making a decision, and I totally respect that. Can I ask you why you were looking into a Tesla in the first place?

Yeah, I mean, I've had my car for way too long now, and so I've kind of taken a look at options, and I have a lot of friends who just absolutely rave about Tesla. And, you know, me, myself, I'm actually a big fan of Elon. Elon Musk. So I figured I might as well take a look and kind of explore my options there and see what you guys have available.

That's a smart approach, James. It's always good to explore your options and see what's available. Out of curiosity, what do you currently drive?

So while it's important to acknowledge the incredible potential of AI to revolutionize various sectors, we must also remain vigilant and proactive in addressing the risks it poses. Balancing innovation with responsible development and regulation is key to ensuring a future where AI works for the betterment of humanity.

Nearly one in five people could have.

Their jobs replaced by AI entirely. That's 300 million people across the globe, equivalent to the entire population of the United States. That's a scary future for many, especially since we're already dealing with rampant inflation and one negative headline after another. But before I tell you that, let me touch on another subject that's also scary for many, and that's the tough economic conditions we have. With high inflation and cautionary headlines, most investors have suffered losses even with a diversified stock portfolio. That's why some of the biggest money managers in the United States have been diversifying outside the stock market into luxury assets with far less correlation. Like fine arts, today's sponsor, Masterworks offers this diversification opportunity, which had once been one of the most exclusive to ordinary investors. Masterworks is an art investing platform. They buy the art outright from contemporary masters like Picasso, Banksy, then qualify with the SEC and offer it as an investment. Net proceeds from its sale are distributed to investors. Since their inception, they've sold over $45 million worth of artwork. And so far, each of Masterworks exits have returned positive net returns to their investors. That includes exits for 17.8, 21.5, and even 35% net returns. Now, historical returns are not a guarantee for future returns. I'm not a financial advisor, and you.

Should do your own due diligence before.

Investing any money anywhere. Like any other investment, there is always, always a risk of loss. Masterworks has over 750,000 users, and there are offerings usually sell out in hours, which is why they've had to make a Wait list. But our viewers can skip the line and get priority access right now by clicking the link in the description you want to use. Bing with images. I realized a secret way to do this. I'm not sure if this is currently being rolled out to everyone, but I do know this is how I managed to access it. So for some reason this, as you can see right here, this is Bing chat. And this is of course, as you know, GPT4 with images and as you know, the capabilities are really, really good. But for some people, they don't have this upload or paste an image from here. So what I did was I googled the Bing chat and for some reason I'm not sure if this is, you know, how it's meant to be featured or how they're choosing to roll it out. But if you go over to Bing.com you might just see this there. Because when I actually went on over to Bing.com and then use the chat in this user interface, I saw the image here, but, but I didn't see it on the sidebar. So I remember a lot of people on our previous video stating that it wasn't released just yet. And of course it might just be rolled out. Maybe it's just my account, but this is something that is worth a try because I did see it here on the bing.com chat but not on the Bing sidebar chat. And of course, as you know, GPT4's images are really, really effective and really, really cool. So it was something that I thought I might as well add to this video. What we had was something from wix. Now if you don't know what WIX is, WIX is a traditional drag and drop website builder which is really easy to use, but now they have this new feature where you can literally just type in what you want and of course build your website. So this is something that we're constantly seeing in business where businesses are being easier in terms of how you can actually create them by using simple AI tools to understand what you state in natural language and then just provide you with whether it be the necessary images. You click one one button and then you're able to generate a website. And the thing about this is that these websites aren't actually bad. These websites are pretty, pretty good. Wix is one of those websites that makes it really easy for beginners to start their blog, maybe to start their new business. And this is what I'm saying, okay? This is why artificial intelligence is about to change many different things in website design. You have to Understand, as someone who previous needed websites built, tools like this genuinely help you out so much. Because for someone to build your website, it genuinely takes around 1,000 to around $5,000 on the normal end in order for you to get that website built. So tools like this not only show, number one, it's easier to build businesses, but number two, maybe some web designers have actually lost their job or some website builders. So this is where you need to pay attention to. Certain things in certain industries are completely changing. But at the same time, it's actually really cool technology that does help out the average person that maybe doesn't have all, all the money in the world to spend on a really custom website. And of course, certain older folks who aren't really that savvy with technology, they can just ask it to build this website and get something that looks as amazing as this. Of course, we had one of my favorite tools. This is what, by far one of the most amazing things I've seen. So this is image 2.3D. And I mean, take a look at what you are actually seeing because this is where you can literally input a single image and get out a 3D print map. Okay? And I just think that this is absolutely crazy because not only have I tried this, looking at this now and knowing how hard it is to generate these depth maps, these bump maps, it's just absolutely incredible how far technology has come. Now, this is very different from other image to 3D. What I'm showing you on screen now is the different image to 3D where they're talking about 3D models, but this is a different kind of 3D. And you can read more about it in the research paper where they try to estimate how terrains are. And it's just really, absolutely incredible. I'm about to show you a live demo of me using this on Google and I'm going to leave a link to it in the description below. So this is the hugging face spot. Essentially it's a demo area where you can test AI tools that haven't been released yet, but you can test them for free. So I grabbed this random image from Google, okay. And I just simply inputted it in. Three seconds later, I got a 3D mesh reconstruction. I mean, look at the level of detail on this. This is literally a random image I grabbed from Google 10 seconds ago. And we can see that I have a nearly highly detailed version of this 3D mesh where I can look at the topography and maybe make estimations or judgments. Now you have to wonder what are the potential applications for this kind of software, maybe you're far away from a scene and you want to really look around the mountains to see if there was something there. Maybe you are trying to look at a crime scene and you want to see how this scenery is going to look or how it works under certain previous conditions. This is truly going to change the game in many different fields and it's going to be really, really effective. I mean, I'm someone who spent some time as well in 3D animating, modeling and rendering. So I really do appreciate how far these tools have come. I mean, look, we can grab this 2D image right here and then of course we can click Submit. You're about to see that in around a couple of seconds we're going to get that 3D mesh reconstruction. And on some images it doesn't work as well, but it's important to note how it does do different images. I think ones where it's like mountains and stuff like that, it does do really well. But nonetheless, you have to understand that.

Once these images and once these tools.

Get a lot better, the ability to recreate scenes in 3D with photorealistic capabilities are going to be absolutely insane. Because usually, essentially, if you're trying to create a scan, you need hundreds of image of a single location.

But with this one image, we're able.

To get a decent amount of information, which means that likely in the Future, maybe just five or six images is going to give us a full 3D reconstruction in high photorealistic detail. And that's something that's not far away. I mean, if you take a look at this image right here, this is the final example I'm going to show you. And then we look at the mesh 3D reconstruction, it just goes to show you what different parts of the image could be. And I mean, that is really, really crazy to be able to look around and have a much more immersive view of that image. I mean, just examining artifacts, examining old images, I mean, it's just, it's just absolutely crazy. I mean, maybe I'm over exaggerating, but I do think that this is something that you definitely should try.

Now this is the most important part of this video.

If there's anything you take away from this video, let it be this short segment that we are currently talking about, okay?

This is a research paper, okay, produced by Stanford University. And in this research paper they talk about how ChatGPT's behavior is changing over time. They talk about the fact that ChatGPT has gotten worse over time. Now, at first I did think this.

Was some kind of rumor that it.

Was happening, but I noticed it in my own example. You can see here that in solving math problems it went from a 97.6% accuracy rate to a 2.4% rate for GPT4, but GPT 3.5 went up from 70.4% to 86.8%. In addition, you can also see GPT4's drastic decline on many different tasks. Now something also that goes on in this Twitter thread by Santiago. He talks about how the team used Chain of Thought to help the model reason is 17077 a prime number? Think step by step and if you've watched the previous videos you'll understand that Chain of Thought is a popular technique that significantly improves answers. Unfortunately, the latest version of GPT4 did not generate intermediate steps and instead promptly answered with a simple no Code generation has also gotten worse. The team built a data set with 50 easy problems from elite code and measured how many GPT4 answers ran without any changes. The March version succeeded in 52% of the problems, but this dropped to a pale 10% using the model in June, Rumors suggest that they're using several smaller and specialized GPT4 models that act similarly to a large model but are less expensive to run. When a user asks a question, the system decides which model to send the query to. Cheaper and faster, but this new approach could be the problem behind the degradation in quality. Now I also noticed this as well because a lot of the time when I was using ChatGPT, when I what you saw in the search bar right here is giving you exactly which model it was. And you can see right here that Davinci 002 was a different model than ChatGPT 3.5 Turbo, and there were many different models that we could see when we went on the ChatGPT webpage. Now, it's not something that was a big story because when I researched this ages ago, like one or two months ago, I noticed that the post only had three upvotes. So it will be interesting to see.

If they manage to update this to.

Reverse the changes or they start to offer many different GPT4 models to many different users in sort of a tier. But this is something that isn't really good for OpenAI's public relations because they've done it without telling people. Let us know in the comments below if this is something that you have personally experienced or something that you are thinking about and your suspicions are now confirmed.

Then we had the United nations talk.

About artificial Intelligence and they managed to bring up some of the key issues that many people are discussing.

They talk about how artificial intelligence could potentially ruin society and have some existential impacts that will affect us all.

I think it's really important to see what our world leaders and people who.

Make the very big decisions, or at.

Least influence them, have to say on the very innovative and moving as quick as we've ever seen artificial intelligence Space.

New technology is moving at warp speed and so are the threats that come with it. Alarm bells over the latest form of artificial intelligence, generative AI, are deafening and they are loudest from the developers who designed it. The scientists and experts have called on the world to act, declaring AI an existential threat to humanity, on a par with the risk of nuclear war. We must take those warnings seriously. Our proposal, global Digital New Agenda for Peace and Accord on the Global Governance of AI will offer multilateral solutions based on human rights. But the advent of generative AI must not distract us from the damage digital technology is already doing to our world. The proliferation of hate and lies in the digital space is causing grave global harm. Now it is fueling conflict, death and destruction. Now it is threatening democracy and human rights. Now it is undermining public health and climate action.

Now, including some AI innovators have called for a pause in the very rapid development of AI technology before it gets completely out of control so that scientists and policymakers can assess where this is going and perhaps suggest some guardrails, possibly including human agency. Do you support such a pause? And do you see the board that you are setting up, including AI scientists, as helping in that process of educating during such a pause?

I think a pause can be an interesting idea, but I don't think a pause will solve the problem because we all know of situations in which there is a pause and then everything goes on the same. I think that independently of the fact that the pause being a positive possible idea, we need to make sure that we move forward and we move forward as quickly as possible in all the other mechanisms that we have been discussing.

Thank you very much, sir. Thank you, Melissa. Thank you.

And we'll see you back in around 12.

James Bayes from Al Jazeera, Secretary General, you've been speaking about AI for quite a long time and I'm sure been thinking about it as well. In your opening statement, you quoted scientists and experts saying that it could be a risk as high as nuclear war. Some are also, though, saying that it could cure cancer. So I'd just like your assessment. What do you think are the challenges and the opportunities of AI?

Well, I think the opportunities are immense. If you look at health, if you look at environment, if you look at education, what AI can produce can be an extremely important factor to make the Sustainable Development Goals a reality. So AI has an enormous potential. But it is clear that AI has a serious problem, which is the possible removal of human agency. And that for me is the central question. It is absolutely essential that human agency remains present in everything that is built with artificial intelligence. But as I said in the beginning, the fact that we look at catastrophic consequences might come from artificial intelligence in the future should not distract us from what's happening today in the digital world that is contributing to people being killed, to human rights be violated, to our privacy being completely destroyed, and for the data that is produced by us to fully escape our control. So many things are happening today that we need to deal with, but at the same time, we need to do everything possible to make sure that the future evolution of AI doesn't go into the logic that completely abolishes human agency and creates a monster that nobody will be able to.