ChatGPT Now 10x SMARTER, AI Can READ YOUR MIND, Open AI's NEW 3D MODEL and More [AINEWS#3]

After another explosive week of artificial intelligence breakthroughs and news, this is going to be a very interesting one because some.

Of the stuff here is actually quite frightening.

So let's take a look at what happened this week. One of the first things that I found that was truly interesting in the world of artificial intelligence was Palantir. Now, this is not something that most people do expect when we look at developing artificial intelligence technologies, but I do think that this is going to become more commonplace as AI evolves and actually becomes a part of our society. You can see right here that in this scenario, AI actually detects that there's some anomalous activity, which just means that there's some strange activity. Then you can prompt the AI to ask it for more details further on. In this example, you can see that the person essentially asks the AI, what military do we have in this region and what commanders do we have that we can send this up the chain and then get this problem sorted, and what would be the most effective strategy to attack or disarm this enemy and neutralize it? And it's honestly very interesting, but at the same time, it's pretty concerning because depending on how advanced this AI gets, we're going to see combat and warfare because potentially play out in a completely different way. So this is definitely something that you should be aware of and something that you should be paying attention to. And I'll show you the full excerpt in the description because it's a 10 to 15 minute video that explains exactly how artificial intelligence will be used in the future to escalate or de escalate certain combat scenarios. The next scenario that we have here is honestly truly terrifying, but honestly amazing at the same time. So it says GPT AI enables scientists to passively decode thoughts. And in groundbreaking study. Now, essentially what this means is that this AI was able to decode people's thoughts by reading MRI readings. And this is honestly something I'd never thought I'd see, says. Researchers at the University of Texas at Austin used FMRI to gather 16 hours of brain recordings from three human subjects as they listened to narrative stories. The team then analyzed these recordings to identify the specific neural stimuli that corresponded to individual words. So essentially we. What they were doing was measuring brain activity when certain words were input. And then of course, you can see right here, this is the actual stimulus that they were given. Then this is the decoded stimulus that the AI was able to decode just.

From the fmri, which is truly terrifying.

Now, one of the main concerns is a privacy concern because of Course, if AI is able to decode human thoughts and essentially know what you're thinking or what you're wanting to say before you even say it, it definitely raises some very ethical questions that we never even thought that we would have to ask. So this is something that is now being brought into because of course privacy is definitely something that it seems like we're losing more and more of each day. And with this advancing technology, there's going to be a lot of different things that we simply cannot predict. Article goes on to state that decoding words from non invasive recordings has been a challenge due to FMRI's high spatial resolution, but low temporal resolution. So essentially, although the images are high quality, the single thought can persist in the brain signals for only up to around 10 seconds, which causes the recordings to capture the combined signals of approximately 20 English words spoken at a typical pace, which is honestly insane. Now the reason this is such a massive breakthrough is before the invention of GPT and large language models, this was pretty much impossible to do. But now that these are fine tuned and now mainstream and everybody knows how.

They work, it's something that is really, really easy to.

Now this one right here is honestly very similar to the previous story that we did just discuss. But this new AI tool uses brain signals to predict what a mouse sees. So researchers were able to extract the black and white footage from mouse's brain signals. So you can see right here that they observed the brain activity of 50 mice while they watched a 30 second clip. Then essentially what they had was they had their trained AI program zebra to link the brain data to the movie clip. And then essentially they reconstructed that movie clip simply from the hidden data recorded in the brain, predicting the complex information that the mice were able to see. And right here you can see the original footage versus the actual footage that the mice were able to see. And honestly, depending on where you stand on AI advancements, this kind of technology did seem like magic at first. But every single day it seems like we're reaching new and new breakthroughs which has pushed the boundaries of what we even thought was possible. And then we had something really cool from Facebook Meta, we had Dynov2, the state of the art computer vision models with self supervised learning. And essentially this is where they have a new method for training high performance computer vision models. So they deliver strong performance and it doesn't require fine tuning. So essentially it's a software that can literally enable you to have perfect computer vision. And the reason this is going to be really good is because this is Open source. Now, why I think Meta are releasing a lot of open source stuff is because they know that AI is moving very, very quickly and a lot of open source projects are being deployed currently. Like a lot of stuff being deployed on GitHub and a lot of stuff being posted on Twitter. There's many different communities where a lot of people are sharing all the advancements. And I'm pretty sure that Meta does want a lot of the new technology to be built on the backbone of some of the stuff that they do create. So it's going to be interesting to see coming over the next couple of weeks how this is integrated to many.

Different products and services.

Essentially now what we have here is one of the most exciting announcements. This is Google's IE and this is essentially their annual conference where they're going to be announcing potentially a lot of things. So here's what's anticipated based on some rumors and based on what previous events have occurred. So Palm 2, which is going to be allegedly capable of handling over 100 languages. And if you don't know what Palm 2 is, Palm 2 is Google's large language model AI. Then of course, generative experiences for search. With Bard potentially infusing Google search with.

AI, this is going to be a.

Game changer because some people are already experiencing and saying that Bard is performing a lot better than it was in its early days. And remember, I've always been someone who's told you that Google does have a lot of AI software that they just aren't releasing. Then of course, AI workspace tools, including stuff that you're pretty much seeing on screen right now. So I honestly am really excited for this event. I do hope that there is some major AI announcements because Google has been sounding the Alarm for what ChatGPT has literally done with Microsoft. So it's going to be interesting to see how the competition does step up. And if Google can reclaim the throne as the number one technology company in terms of innovation, then of course we had OpenAI actually release their Shap E. So essentially what this is, this is text to 3D text 2 nerf and essentially why this is a big deal. Despite these small images looking like they're not much, essentially their models can generate complex and diverse 3D assets in just a few seconds. The text says here, though the created 3D objects might look pixelated and rough, the models can be generated with single text. And another limitation that's included with this currently is that it's only capable of producing objects with single prompts and simple attributes and can't do many combined objects, as pointed out in the paper. Now, of course, remember this is still early, so for people that are going to say it looks terrible, remember what Midjourney V2 looks like. And then of course, what we had a year later, and it also does.

Look like OpenAI had released point E.

Earlier before, and it was essentially just.

A 3D version of Dall E. So it's definitely going to be something that I do think they're going to be working on in the back end, because text to 3D is definitely something that these companies are trying to master. Then, of course, we had been getting a major upgrade and one of the things that is actually happening now is you now can have chat history, which is really, really cool because Bing usually only allowed you to have 15 messages or around five to 10, which was very limited. Of course we had another feature of the upgrade actually allowing Bing integration plugins, which are apps that Bing can then use to make your life more easier. So for example, if you ask it to book a restaurant, it can actually go ahead and use the OpenTable app plugin to book that for you. Then of course, we had some other stuff, such as Bing actually having more features yet again with the new multimodal features that are going to be integrating Bing Image creator, which previously only worked in Bing chat, if you selected the creative one. And I've seen many people get responses where you actually have images in chat now, and some of them even actually make graphs. And there's just many different things that are making this chat honestly much better than ChatGPT in certain aspects. And of course you can see behind me these images in the trailer, you can see downloading the text, you can see copy and pasting that text. And, and essentially what you're seeing is the complete version of Microsoft's Bing Image creator, all integrated into Edge, allowing you to have a much more smoother user interface. And I guess you could say overall experience and it's really, really good. And something really cool is that in cryptocurrency we're now starting to see AI crypto powered bots. So you can see right here, crypto.com unveiled Amy, which is an AI powered crypto computer companion. And you can see right here that it's essentially designed to help users learn about crypto and give them real time information about the markets. And essentially what these things are designed to do are essentially just designed to educate you. Because I know that many people might be thinking that these are going to be giving people buy and sell signals and telling them exactly where the market's heading. This is not what they're trying to do because I've used these bots and that's not what they're like. You can see right here it says Amy does not provide financial investment advice and this is in its beta. But that's what they're designed to do.

Because if these AI chatbots do provide.

Investment advice, they essentially might hold some risk. And that is something that these companies really don't want to have. You can also see that Binance here actually added their AI driven chatbot, which is Binance Sensei, which is driven by ChatGPT's technology. And I'm pretty sure they just trained it on a fine tuned large amount of financial crypto data, which means it's going to be geared towards helping answer those questions which are very crypto related.

Which is definitely going to be very.

Useful for those who are getting into the space that have a ton of questions. Because crypto is something that just like AI is evolving pretty, pretty quickly. Then of course we had something more on the creative end which was really cool. And I found this text to 3D Game World software which is called Blockade Labs. And essentially you can create, transform any landscape into any kind of theme that you wish simply with a text prompt. So the current footage that you're seeing from Blockade Labs shows the many different environments that you can create based on a certain text prompt. Provided you have the right 3D environment. It's actually really, really cool. And you can actually test it. Right now you can see right here where I simply prompted something. Essentially I tried to get New York City, but it was more tuned to, I guess you could say, a fantasy esque theme, which is essentially what I managed to get in the end. And then I also did a second prompt in which I tried to get a hellish underground themed area. And then of course you can see right here that it's. It still kind of went to that fantasy area. But I still think that this software is going to be very, very good in the future if you're able to potentially, you know, use this in VR or use this in some interactive kind of experience. Because the quality of this was honestly really, really good. And I still felt that it delivered on what the prompt was. Then we had something really interesting was where the President of the United States actually spoke about AI. So it just shows that this is a thing which is honestly starting to catch the attention of the policymakers around the world. And essentially you can see right here where Biden sat Down with some of the top AI CEOs to discuss AI regulation and AI safety because that is honestly something of everyone's concern right now, considering the rise of AI is moving very, very quickly. I'm pretty sure that these policymakers and these lawmakers want to ensure that there aren't any regulations or any sort of rules being broken. Because of course AI definitely could pose a risk to the public if it does get in the wrong hands or if there is some kind of runaway type scenario where it bluntedly does something that we didn't expect or it just plainly does something wrong. Now, essentially some very, very interesting News was that HuggingChat is a new competitor to ChatGPT and I'll walk you through a live demo of this tool because honestly it's genuinely is really quite impressive for such a small scale model that can literally be run locally. They're using a fine tuned version of Llama, which was Facebook's Meta's AI language model which was leaked onto the Internet that many people have fine tuned. You can see right here that they have two models open assistant llama, the 30 billion parameter large language model. And you can see right here this is the live demo of the tool. And here honestly it is very, very interesting as to how it responds to the questions. I would argue that in certain aspects it's actually a lot better than Bard. We know that Bard suffers from quite a lot of hallucinations, but to be honest with you, for a very small parameter large language model and for one that potentially can be run locally, because there are instances of people actually running these locally. And what I mean by that is people that are running these on their home systems, so just tuning it so that it can run on their computer, it is very, very effective at small scale tasks, which is going to be interesting considering that one of the large limitations of large language models and products like GPT 3.5 and GPT 4 is of course their cost to run and cost to maintain. If we can get these large language models down to a small enough size where we can run them very efficiently and effectively on our local computers, then that will literally change the game. So here you can see it answers certain questions. For example, this question was to how to improve my YouTube channel and it definitely gave me a much longer and much more extensive report than ChatGPT did, which I found to be very interesting as well. So the token size window for this large language model is actually quite high, surprisingly, and it definitely did impress me more than I would have thought. Then of course we had a video by Dr. Alan D. Thompson, which was very, very telling because he spoke quite a lot about theory of mind. And theory of mind is something that we never essentially intentionally trained AI to learn how to do, but it just came as an emergent capability. And essentially what that means is that as the capabilities of the AI grew and as the parameter sizes increased and as it got smarter, this theory of mindability became more and more apparent. And essentially what theory of mind is, is the AI, or the large language model, can essentially understand or predict what you're thinking and how you're feeling and essentially understand your mental state, even though it isn't really meant to. So based on the information that you give it, it can literally understand what you're thinking. And the reason this is very crazy is because it pretty much hit the ceiling on the theory of mind tests, which means that it's far better at humans at thinking in terms of empathy, understanding the emotions of another person. So it's really crazy understanding what's going to be going on in your brain.