AI's Most Insane Week - Things Will Never Be The Same

So in all of the time that I've spent messing with AI and following AI news, this week has been the absolute craziest week in AI that I've ever seen. Every single day there's been some sort of crazy announcement going on. And in this video I just want to break down this week in a nutshell. Let's get into it. To start off on the week Monday we got news that Stanford introduced the Alpaca 7B model. What this is is a model trained on 52,000 instructions 6 similar to the way the GPT3 model is trained. However, because it's trained on so much fewer instructions, it's much more lightweight and can actually be used on somebody's local computer. When tested against tools like GPT3 and ChatGPT, it actually performed almost as well as those models, but with such a smaller file size and less necessity for compute power so that it sort of democratizes these large language models so that anyone can just run them on their own system. Elvis here on Twitter browser breaks it down by saying large language models are getting cheaper, better and more accessible. Now they tested this on a really, really high end graphics card and it took three hours to train it and they fine tuned it with the Llama models which was basically recently released to the public. Carlos Perez here says, I don't know what to make about this development. Alpaca is surprisingly very good. The claim here is that training can be done in five hours using a single RTX 4090, a still fairly high end graphics card, but a consumer graphics card. And then he asks, has GPT like models been democratized overnight? Now, while most of probably won't be trying to run our own large language chat models on our own computer with what's currently available, what we're seeing with Alpaca shows that we are really, really close to this. It won't be very long where similar to stable diffusion, you'll be able to just install your own chatbot on your own computer, have it trained on your own data set and have Your own personalized GPT3 or ChatGPT on your own computer without even needing to be connected to the Internet. We are very close to that and Monday was a big breakthrough and one step closer to that. Now on Tuesday, Tuesday was a huge day. So much happened on Tuesday. Everything just kept on getting overshadowed by everything else. Now to start off, Google announced that they were going to be releasing all sorts of AI functionality inside of their workspace tools. We're going to get GPT3 like functionality inside of Google Docs and tools like Gmail really soon, where it will just auto complete your emails, autocomplete your docs for you, help you brainstorm ideas, write entire articles for you, write your resumes, write your cover letters. It sounds like it's going to be pretty much what we expect from tools like GPT3 directly inside of Google Docs and Gmail. And then rolling out a little later after that, we're going to start to see it integrated inside of Google Sheets, inside of Google Slides inside of Google Meet, and all of the various other suite of tools that Google has out there. We're going to start to see a lot of this AI roll into those tools is probably going to make a lot of these recent crop of AI based companies fairly obsolete since a lot of these companies are actually building tools that are kind of just going to be features inside of what Google is building now. Also on Tuesday, Google announced the next generation of AI for developers in Google Workspace. They announced that they're releasing the Palm API to select developers so that they can start building on top of the Palm model. Now I recently did a video where we talked about Palm and Palm is a multimodal model where it's going to be able to read what's going on in images, it's going to be able to read your text prompts and combine that with images. It sort of does a lot of what people were expecting GPT4 to do, and now a lot of companies are going to get to build on top of that because of the API that's getting released from it. Also on Tuesday, Anthropic, a company that Google themselves is heavily invested in, introduced Claude, which is their own version of a chatbot. At the moment it appears that Claude is only available through API. So Claude is being used behind the scenes on a lot of tools like PO from Quora and like the AI that's built into Notion. And then on Tuesday, the granddaddy of them all, the biggest news of the week, GPT4 is ready for us to use and it was in Chat GPT for us to play with that same day. It was a little interesting because chat GPT didn't even know it was on GPT4 on day one. But the outputs that everybody was getting when messing with GPT4 inside of ChatGPT was clearly different and better than what we were getting out of previous versions of ChatGPT. Now GPT4 also announced that we're going to be able to have longer context in our messages with them and that it's also going to use visual inputs and be a multimodal model as well, where it will be able to look at images and use that image for additional context and decipher what's in the image and do all sorts of interesting things with the combination of image and text inputs. This was a huge breakthrough on Tuesday and really set the AI world on fire by the outputs that people were getting out of GPT4. So much better quality of a responses, but the accuracy of the information that it was putting out, it was just so much better. And then on the same day, Microsoft then confirmed that for the last five weeks, if you've used Bing at all, you were using GPT4. They said, we are happy to confirm that the new Bing is running on GPT4, which we've customized for search. If you use the new Bing Preview at any time in the last five weeks, you've already experienced an early version of this powerful model. So it wasn't what we see today with GPT4, but it was an early version of GPT4. We have been using GPT4 in Microsoft Bing, like I stated in the very first video that I made when I first talked about Microsoft Bing chat. All right, so we're through Tuesday. Now then on Wednesday, Midjourney was the talk of the day. They held a live office hours and 2,000 people show up to the live office hours because on their Twitter they. They announced that they were going to make an announcement and everybody kind of knew what that announcement was gonna be, but they actually made two announcements. The first announcement was the announcement of the new Midjourney magazine, which is a curated collection of images from the community as well as interviews and content around generative AI. They also announced that you can get access to Midjourney magazine and get your first issue for free by clicking on get the magazine and then adding the promotional code subscriber and it will bring your total down to $0. So you can get your first month free of Midjourney magazine. Now that was pretty cool. It was a nice announcement and everybody was excited about it. But the real big announcement was that Midjourney just launched Midjourney version 5. And Midjourney version 5 was able to create much more realistic images. It was able to do a much better job at hands. Although still not always perfect, the images are much more photorealistic. They have a new tiling feature that you can use to make tiled images that line up perfectly with each other if you use the image over and over again. They added image weights back into this new version and the way you prompt is completely changed. Now they want you to prompt in a way where you're using full sentences and talking to it like you would talk to a chatbot, like ChatGPT. So it's got much better recognition for normal standard language. And they also revealed on this call, towards the end of the call, they didn't go too deep into it, but they did reveal that they're working on an API that if you want to get access to the Midjourney API to get in touch, which means we're going to start to see other products start to pop up, most likely that use Midjourney for their art generation. It's also very interesting that the way that you prompt changed with Midjourney version 5 to more of a normal sentence structure where you would write like you would talk. But that also fares well for connecting it to things like GPT4. GPT4 is going to output more like sentences and less like what we were used to prompting in Mid Journey. So this could mean that it will be a lot easier to allow some of these chat tools to generate prompts that Mid Journey will easily understand. So that was the big news on Wednesday was Midjourney version 5 is huge and it generates some really, really amazing images. And then we jump to Thursday and Thursday Microsoft announced their 365 copilot. Similar to what I mentioned about what Google was doing with their workspace. Microsoft is basically adding in AI to everything that they do as well. So pretty soon we're going to get AI inside of Microsoft Word, we're going to get it inside of Excel, we're going to get it inside of PowerPoint, inside of Outlook, inside of Microsoft Teams. And they also introduced their new Business chat and that takes all of your Microsoft tools and uses the data in all of these places to answer chat questions. So essentially, if you've got Microsoft Word and you've got Microsoft Excel and you've got PowerPoint and you have Microsoft Team meetings and you and it summarize notes for you and you've got Outlook, it basically takes all of the data from all of these places and lets you chat with it and it will look in all of these places to answer your questions. So do I have any meetings on Tuesday? It could look inside of your email and see if there's anything that comes up for Tuesday and answer your question. Remind me what we talked about on that meeting last Thursday. It could look at your meeting notes and pull in data about what was on that meeting. It basically creates a fine tuned chatbot for you of all of the information that's inside of your Microsoft suite of tools. Also On Thursday, the ChatGPT rival made by Baidu called Ernie was released and it was off to not the best start. They gave a one hour presentation on it and well, everybody was pretty underwhelmed by it. They didn't show it off live, they pre recorded all the responses to carefully curate how it was going to respond and because they didn't really want to show off what it can do live, and they sort of carefully curated how it was going to respond in prerecorded messages. The company's shares of the Baidu Stock slumped by 10%, so shareholders were not impressed either and it was kind of a miss. It doesn't look like they're going to be competing real tightly with Google or Microsoft anytime real soon. And that's just through Thursday and I'm recording this video on Thursday, so if there's more big news on Friday, I didn't capture it in this video because of the timing of this video, but next week it's not going to slow down. Next week is crazy as well. Next week kicks off the GTC event, which is Nvidia's conference all about AI. It runs from the 20th through the 23rd of next week and Jensen Huang, the CEO of Nvidia is speaking on March 21st. As you can see, they describe it as the conference for the era of AI in the metaverse, and who knows what kind of big announcements are going to be made during this conference. You can go to Nvidia.com GTC and it is free to watch it virtually online. There's going to be a lot of presentations about the latest research and how all of this technology is going to transform the world. And there's people speaking from companies like DeepMind, the co founder of OpenAI, researchers from Meta, the chief Strategy Officer from Adobe, and a ton more. So it's going to be exciting. I want to be trying to tune into this event every day and if you know me by now, you know I will probably be trying to make videos about any cool breakthroughs that come out of this event. But next week is going to be crazy and I'm not even done yet. Here's one little thing that I'm going to leave you with before this video ends. Imad, if you're not familiar with him, he is the founder of Stability AI, the company that created stable diffusion on the 15th. He said lots of announcements. The next few weeks will be fun. As an aside, even after the Amazing Midjourney version 5 release today, I don't believe the releases not from us over the next week are even done. What a time, eh? He's basically hinting that there's some big announcements coming next week about Stable Diffusion, and there's definitely some things in Stable Diffusion that are in the works. We're expecting a stable diffusion 3.0 model to be announced sometime real soon. There's Stable Diffusion xl, which has been in talks. There's talks of an overhaul of the Stable Diffusion online user interface. So who knows what announcements he's going to make next week. Emad is someone to pay attention to on Twitter if you're not already, because he's one of the guys that's at the forefront of the AI space and he's telling us right now that these big announcements that are happening this week, they're going to continue into next week. If you want to stay on the pulse of AI, make sure you subscribe to this channel. Make sure you like this video to make sure you see more of them in your newsfeed. And if you haven't already, head over to FutureTools IE, click the button to join the free newsletter and every Friday I'll send you the TL Dr. Of the Week in AI. I'll tell you about my five favorite tools, all the most interesting news articles that have happened in the week. I'll share some YouTube videos with you, and I'll tell you one cool way to make money with AI. And I send that every Friday. And all you gotta do is go to FutureTools IO and click the button to get on the list. What a week. What a week. This has been the most intense week in AI that I have seen yet and man, it's been fun and exciting and I can't wait to see what next week and what the next months hold because this stuff is just picking up speed fast and I'm here for it. Thanks so much for tuning in. I'll see you guys in the next video. Bye bye.