Snapchats New AI, Elon Musks New AI, GPT4, AutoGPT, , Facebooks New AI [Weekly Dose Of AI #1]

So let's take a look at everything that was released last week in AI. And boy, oh boy, pay attention because there is a lot of stuff that you missed. Coming in at number one, we have GPT4's multimodal feature being, I guess you could say, expedited by Microsoft. Essentially, the researchers at Microsoft decided to release Visual ChatGPT where you can actually use ChatGPT to chat with certain images. Now, the videos that you're seeing on screen are essentially demos and there is of course a research paper in which they describe exactly how they use ChatGPT to, to describe and chat and of course generate these images. Now, of course, you can actually use this if you want. The web page is still live. All you're going to need is an OpenAI key. And once you simply put that in, you're going to be able to simply use the web page and interact, just like I did. Now, Bloomberg also released their large language model which is going to be centered around finance. Think of it like ChatGPT, but essentially just specifically trained to look at finance models. What they did was they trained this large language model on all the finance papers that they ever looked at. And essentially the resulting language model that they got was one that can accurately predict and make certain guesses about market sentiments. And over time they're going to be improving this software by adding more and more parameters. They had Facebook releasing their new Segment Anything model. And essentially what this Segment Anything model was, they use AI to detect every single item or object that might be in an image. And now this isn't just insane because they're detecting absolutely everything that's in an image. They actually know what the item is in that specific image. So it's an image identification and a classifier. So right here you can see that it's able to identify how many cats are in this entire image, which is going to be good. And has a variety of many different applications worldwide. We also had them showcase exactly how Facebook's segment is going to be used in a real world scenario. So right here you can see that these AR goggles that are being deployed on this user are able to accurately identify whatever is in the field of view. So many people do have trouble seeing things, such as those who are partially blind or those with visual impairments. Maybe there are also some who see a new object and they need to get that updated into their interface as quickly as possible, as maybe there are some foreign objects or some specific objects that are very rare to find. This could host a whole world of many different applications. And it just goes to show how quickly this is being used and how it can actually be used for everyday scenarios where maybe you want to sort something, maybe you want to find something quickly. Now what's also cool is that they also showed us what their plans are for the future. For example, they showed us exactly how they plan to integrate this into everyday cooking activities, such as here. When you're making a new meal, you might be confused on the cooking steps. So it actually is able to identify exactly what items you need, how much they are, and how you can actually use them effectively. We're also able to see maybe if someone is forgetful about feeding their dog, you know, this is going to be something that is very useful for people who do have potentially some disabilities, where they don't remember everything that was going on in that particular environment. Something that's really, really useful. Then we had Microsoft release their new software, which was Jarvis. Essentially what they did was they combined ChatGPT with HuggingGPT, and essentially they managed to get ChatGPT to complete pretty much any tasks that a user can ask, even if it is multimodal. So of course the first request was please generate an image, whereas a girl is reading a book and then make sure her pose is the same as the image that I've just submitted. And then describe this image with your voice. And of course you might be thinking, how on earth is, is ChatGPT or Jarvis going to do this because it doesn't have a voice. You can see right here that the four stages then are commenced based on all the tools that JARVIS has access to. So we had stage one, which was planning. You can see right here that chatgpt JARVIS is managing to plan out exactly what it is going to do by selecting everything and the sequencing of orders. Then of course we have the model selection where the JARVIS AI decides, okay, I'm going to use this specific model because it needs to complete this specific task. Then after stage two, what we had was of course stage three, which is the task execution where it uses the AI softwares, whichever it may be. And then of course it has the response generation to generate the desired output. So definitely something that was really good. And of course the final response was there, as you all can see on screen. And JARVIS proved to be a very good success. And it actually documents how it does this. So it definitely proved that Microsoft are still, every single week they're getting better and better at utilizing the AI to make different requests able to be done. And of course you can see right here, they're able to identify exactly what is going on in the image. And remember, this isn't ChatGPT4. It was just something that is done with a lot of the tools online. Now, of course you can actually use Jarvis if you want, you can go over to the website hugging GPT and all you're going to need is some API keys and you're going to be able to use that website. Bear in mind it is a little bit buggy, but it is something that you can use. Now, something that I thought was very, very interesting was the deployment of some robots to help patrol Times Square in New York. Take a look at this. Announcing three new policing technologies in New York City. The K5 autonomous security robot, the Spot Digi Dog Robot and the StarChase GPS attachment system. Our job is to fight crime and keep people safe and these tools are significant steps forward in that vital mission.

We are here to get stuffed up gst.

We also had someone showcase exactly what's possible when you provide Gen1 by Runway with a perfect driving image and a very good video reference. I do want to say shout out to this person that created this because it was definitely very, very creative but and just goes to show exactly what's possible with a perfect combination of the artistic styles used to use text to video and definitely shows what's going to be capable in the future. Then we had Amazon releasing their new software which is called Bedrock and essentially allows people from all around the world, businesses alike, to access a large model of AI software which they can easily use and fine tune to their business specific needs. Now you about to see a clip of the CEO talking about why they launched this and how effective it's going to be.

Companies don't want to go through that. And so what they want to do is they want to work off of a foundational model that's big and great already and then have the ability to customize it for their own purposes. And that's what Bedrock is, which we now say, which is it'll give you access to large language models from Anthropic, from stability, AI from AI.

And you can see right here that this is on Amazon's Bedrock's new page. You can, you have text generation, you have chatbots, you have search, you have text summarization, image generation and personalization. And you can all see right here that you have the choice of the foundation models that you can use that is going to be available on the Amazon Bedrock platform that you can then fine tune for your business specific needs. What they also did release as well was CodeWhisperer, which is essentially an AI coding companion which is going to help people get more done faster. So definitely another area in which Amazon is deciding to innovate. Then of course we had Nvidia release in its Text to video paper in which they showcase many different examples of how they manage to use stable diffusion to then generate some very high fidelity videos that are quite temporally consistent. You can see right here from all of these examples, this is exactly what they've been able to do. I'm not sure how long they've been working on this, but a lot of the footage that I have seen does look far better than some of the previous examples that we have had from the other companies that have tried before. And text to video is by far one of the most hardest things that we are looking to see when it comes to generative AI. And you can see on screen right now that some of these highway examples are very, very good. And some of these landscape examples are also looking very, very effective as well, as they seemingly manage to stay pretty smooth and pretty coherent in what the animation looks like. So I think Nvidia definitely are moving in the correct direction because because they seem to have a step ahead in front of where everyone else is at this moment in time in text to video. Then of course we had Google's own Text to Video where you had input images being able to generate a fully fledged video. A lot of these videos do look very, very well. I mean, the difference between this and Nvidia is honestly quite interesting because they both use very different ways to generate the result. This one uses a multitude of images, whereas Nvidia's is mainly based on text to video. But they can do exactly what Google is doing here, where you input a number of driving images and then essentially you do get an output result where of course your driving image can actually be replaced or placed in a random environment. Once again, Google showing how far ahead they are in the AI race and that they're actually not that far behind of their competitors. And of course we had Elon Musk who decided that he was going to create his own Truth GPT because he realized, and many of the users who use ChatGPT realize that it is somewhat inherently biased towards certain arguments. For example, right here, this article was Talking about how ChatGPT's inherent bias was particularly focused on certain topics and declined to respond when asked certain questions about certain topics. And then of course, Elon Musk has decided that he wanted to create a ChatGPT version that would pretty much do anything the user requested, as long as it's within the standard ethical confines, but just reporting the data.

A path to AI dystopia is to train AI to be deceptive. So yeah, I'm going to start something which I could call Truth GBT or a maximum Truth seeking AI.

And he also talks about why it's not really about politics, but more about.

AI safety that tries to understand the nature of the universe. And I think this might be the best path to safety in the sense that an AI that cares about understanding the universe is unlikely to annihilate humans because we are an interesting part of the universe. Hopefully they would think that, I think because yeah, humanity could decide to hunt down all the chimpanzees and kill them, but we don't.

Then of course we had Google announcing that it's going to be launching a new AI powered browser with which is going to be integrated into Google called Project Maggie. And essentially this is going to be a direct competitor to Bing's recent launch of ChatGPT within the browser and they're going to try and expedite this as quick as possible and release it next month, which is actually going to be sometime in May. Now Google decided that they needed to get this out as quick as possible because recently Samsung and Apple were in talks of actually switching up browsers in which that they were going to be having as the default option on their phone. So Google was pretty much forced to start developing this as soon as possible. Then of course we had the release of AutoGPT, which is now powering Agent GPT, which is essentially a form of AI which you give it one prompt and then it runs off and decides to organize itself with a bunch of tasks and then it starts to scale the Internet and essentially execute on those specific tasks. Now this has absolutely blown up on the GitHub and, and this is something that people are now using to generate a bunch of different texts, a bunch of different articles and complete many tasks. And I do think that this is likely what the future of AI is going to be because it removes a lot of the work that people initially will have to do depending on the nature of the tasks. And many people are now discussing the fact that once you get these certain agents to be able to do certain things, companies are going to hire these agents for, for cents on the dollar. And then of course we might actually see some major layoffs coming. Then of course we had Quora release their new AI chatbot which is called po. Now, what's good about PO is that you're able to customize different chatbots to your liking to fit a certain style or character, then talk with that chatbot in any manner possible. What's Also cool about PO is that they also give you access to GPT4. They give you access to Claude Instant, which, which is another competitor to ChatGPT. And as you can see right here, I was setting up bots that allow me to interact with someone in Steve's job's personality for advice on YouTube. And then of course they have the original Sage bot, which is a general knowledge bot, which was released by themselves. Now, of course, what we also had was we had Snapchat release their own AI bot, which is essentially my AI and it's called their Sidekick. But if I'm being honest, the results that I've seen floating around the Internet are very, very up and down. Some people report great responses, while others report responses that just simply aren't even that accurate at all. But I'm guessing that they wanted to rush this out pretty quickly, which is why I guess we're getting this in its unfinished sort of state. Of course, we had, you know, the Bard, which was initially rumoured with bugs and many mathematical failures, and getting a very nice update, which is going to help it be powered with some of the processing power from Palm, which is the 540 billion parameter model, which is very effective at using robots to do many in real world tasks. And this should make Bard much more effective at the tasks that it does.