An Actually BIG WEEK In The World of AI

Well, I couldn't be more excited because AI news is really, really ramping up. January, quite honestly, fairly slow month for the world of AI. This week we've got announcements from Google, from Microsoft, from Apple, from YouTube, from Midjourney and so much more. A lot is happening right now and if we thought last year was a wild ride for AI, well, it feels like this year is going to be no different. So let's break it down, starting with the fact that we just got got Gemini Ultra. Google is completely moving away from the Bard brand and going all in on Gemini with all of their AI products. In fact, if you go to bard.google.com it just redirects you to gemini.google.com now. And on the new Gemini we have two options. You can see we've got Gemini and now we have Gemini Advanced. Most likely you'll see a button to upgrade, but I've actually already upgraded to Gemini advance because, well, I, I published a video yesterday all about Gemini Advanced and Gemini Ultra. So I don't need to go too deep into the Google News on this one because we did a pretty deep dive here in yesterday's video. If we take a look here at the updates page on Gemini, you can see Bard is now Gemini. And the new Gemini Advanced, which costs 20 bucks a month to use, uses the Gemini Ultra 1.0 model, which supposedly is the model they showed off a few months ago when they made a big deal about Gemini Ultra. And they and then we found out kind of the next day that most of that video was sort of faked. But according to the press release here, it's far more capable at highly complex tasks like coding, logical reasoning, following nuanced instructions and creative collaboration. It's available in 150 different countries. It currently has an Android app with an iOS app in the works. However, if you watch the video that I made about Gemini already, Gemini Ultra slash, Gemini Advanced hasn't really lived up to the hype yet. It's really fast and it's pretty good at creativity on par with GPT4, but then with problem solving and coding and explaining what's in an image and even generating images were all pretty overwhelming for this being the model that Google claimed was going to be way better than GPT4. But if you want to check it out, I highly Recommend going to gemini.google.com, clicking to upgrade to Gemini Advanced and you actually get two months free. It's 20 bucks a month. That's after the two month trial period. So if you want to give it a go go. Try it for two months free. Mark your calendar. If it's not doing it for you, just make sure you cancel it before the two months is up. And you'll never spend a dime on Gemini. But definitely check it out. Compare it to chat GPT yourself. So far, personally, I still think ChatGPT is a little bit better, but I do fully expect Gemini to improve at a very rapid pace at this point. And then and the timing of this Gemini launch is pretty interesting because it comes right on the heels of Microsoft making some big changes and updates to Copilot. And it also comes several days before the super bowl where Microsoft is planning to release a Copilot ad showing to the world what it's capable of. I do have a video next week coming all about Copilot, where we're going to dive in and talk about a lot of the new features that Copilot has rolled out. However, one of the most impressive features that I've noticed recently is with their Image generation tool. If we go to co, you can see we have a redesigned homepage here. It looks slightly different, slightly cleaner, but still gives you some prompts that you can test out to see what it's capable of. But if we tell it to design an image, there's some new image features they've rolled in. I'll get a prompt like create an image of a humanoid robot on a cliff watching the sunset. While we're waiting, we actually get this really cool animation. And once it finally generates our image, check this out. If I click into any single image here, like this one, and then I move my mouse around over the image, you can see it's automatically segmenting the image. And I can select any piece of the image. If I keep clicking, I can select more of the image. So I can select this full robot here and then it gives me two options. And I'm sure there's going to be more options coming really soon, like the ability to remove objects altogether and possibly even inpaint what we're selecting. But right now we get Color Pop and Blur background. I click Blur background. You'll notice it made the background slightly more blurry while the robot is still in focus. Let's go ahead and undo it. You can see it gets a little more clear. I'm going to go ahead and select our robot again and this time let's try Color Pop and you can see what it did. It made the whole rest of the image black and white while leaving just color on the robot. Let's go ahead and undo that. And this time let's just select the ocean and the sky and do a color pop on that. And now we've got a black and white robot. And the hills are black and white, but the sky and ocean are colorful. Another really cool feature about the new image generator right inside of Copilot is these restylization options. I can select low poly here and it will essentially generate the same image but in a low poly style. Here's what it looks like if I use watercolor, here's what I get when I do pixel art, and here's what it looks like in claymation. The Dall E3 art generator directly inside of Copilot as well as inside of Microsoft Designer has gotten a huge upgrade. And I'm really impressed with what this can do. I mean, it's blowing away what Imagen instead of Google is capable of right now. For today's video, I partnered with Tubebuddy from Ben Labs, a company that I've been using their tool for a long time, long before they ever sponsored me. And now they've even got some really amazing AI features built into the platform. So check this out. If I log into my YouTube account and go to edit one of my videos here, you'll notice a new little TubeBuddy icon on the right side of my dashboard. If I click in here, there are so many cool things I can do with my videos. One of the really smart new AI features is this suggested shorts feature. I simply click this. It will then analyze my video and then tell me exactly which chunk of the video would make a really good short. So this suggests 2641 through 2656. Not only does it suggest shorts for this video that I'm on right now, it also gave me some suggestions from other videos on my channel that would be good candidates to create shorts from as well. My other favorite AI feature is the title generator. You'll notice a new button showed up right below where I enter my title. And if I click on it, tell it a little bit about what my video is about and click generate ideas. It will use AI to give me a handful of additional potential title ideas. You probably won't be shocked to learn that most of the titles that I use on this YouTube channel AI helped me generate. It's only fitting, right? And TubeBuddy gives us the tool built in right into our YouTube dashboard. So I don't need to go to any other third party apps outside of YouTube. I can get exactly what I need inside the platform again. I've used TubeBuddy for years for my own YouTube channel. It's one of the ways that I've really managed to grow this channel. So of course I couldn't have been more excited when TubeBuddy offered to partner with me and sponsor some of my videos. So thank you so much again to Tubebuddy. If you're using YouTube and you don't already have a Tubebuddy account, I'll make sure the link is in the description, but it's really gonna help you make life a lot easier when trying to grow and scale your online presence with YouTube. Since we're on the topic of AI art already, let's talk about Mid Journey, because Mid Journey has some major updates they just announced, including the announcement that if you've generated more than a thousand images on Midjourney, you can now actually use the Mid Journey website to generate your art. You can finally get out of the Discord to generate images. Assuming you've generated over a thousand images with Midjourney, you can Simply go to alpha.midjourney.com and it will bring you to this page where you can generate your images. You can see up here we've got a little box that says Imagine. And just like in Discord, we typed whatever. We want our prompt to be here. But instead of adding a whole bunch of extra strings and commands to the end of our image, I can click these little settings here on the right side of the search box and we can change to portrait, square, landscape, and pretty much all of the various aspect ratios that are the most common. We can adjust the stylization, the weirdness, the variety. We can set it to standard mode or RAW mode. We can generate in any one of the old models or the Niji models, and you can set your speed to Relax, Fast or Turbo. Since I've already used this prompt already, let's go ahead and use it again. A humanoid robot sitting on a cliff watching the sunset. Let's set it as a 16.9 aspect ratio. I'm going to crank up the stylization a bit, crank up the weirdness a bit, leave it on standard version 6 and fast. And if I just hit enter and if I'm on this Create tab over here on the left, I can actually see the percentage done that it is. It hasn't started yet, and after a few seconds we get some images that look like this. Now, hovering over the images, we have the option to very subtle or very strong. Over on the right we have options to rerun, edit, hide, and under more we've got Copy, report, download or open. In Discord, if I click on Edit, you can see it puts the original prompt back up here so that I can just create a new prompt if I need to. I click on an individual image like this one. It gives me a whole new set of options. Over here on the right we've got very subtle, very strong upscale, subtle upscale, creative remix, subtle remix, strong pan up, down, left, right. The various zooms change aspect ratio. We can rerun, vary region or reuse the image or the prompt. If I vary region, it takes me to a page where I can just select a certain area, select this area and put add a baseball cap, it'll submit the job. I click on Create, I can watch as it gets completed here and we have some images where it attempted to add a baseball cap. So pretty much everything you can do in the Discord seems like you can do directly on the website now. However, there is one feature that recently got rolled out that I don't know how to do on the website yet. I only know how to do it in the Discord and it's a feature called Style Reference. This helps you get a consistent style across every image that you generate. Let's say I really like this pixel art image here. I could take this image, drop it into Discord and go ahead and just save it in my Discord. Now that I have the image uploaded, I can pick a prompt and prompt it in the same style. So let's go ahead and type. Imagine a humanoid robot sitting on a cliff watching this sunset. But then I add this additional code at the end. Dash dash sref that's short for style reference. And then right after the style reference, I'm going to put a space and then I'm going to copy the link to this image. So I'm going to right click Copy link right next to our SREF here. I'm going to paste our URL here and it should use the above image as a style reference for our new prompt. And just like that, we got some images of a robot on a cliff staring at the sunset. Pixel art, same color palette and everything. A really cool feature if you want to make sure you get the exact same style to every image that you create. Now there are some other options and ability to blend multiple images together and set weights and all sorts of stuff like that. So something to play around with and make some really cool images. In mid Journey, Apple's also getting in the AI image generation game. They're working on a new model called MGIE or MLLM guided image editing here's the actual paper on archive.org where they show really what it's capable of with the types of text prompts you would give it. So here's a picture of a pepperoni pizza. They gave it a prompt of make it more healthy. It then spits out a new pizza with extra vegetables on it. Here's an image where there's somebody in the background here. They gave the prompt remove the woman in the background and then the woman is removed. It looks like the couch she's sitting on is removed as well. Here's another image where they said add contrast to simulate more light and it brightened up the image, let the laptop have a green web page, and it changed this image to this image. There are a whole bunch of other examples inside of this paper that really show you what it's capable of. But this is new research coming out of Apple that we don't necessarily have access to, but will probably be in Apple devices fairly soon. And since we're on the topic of AI art generation, Both Meta and OpenAI have talked about labeling the images that are created with their various AI art generators. We can see here that on this Meta article in the coming months we will label images that users post to Facebook, Instagram and threads when we can detect industry standard indicators that they are AI generated. They've labeled photorealistic images created using Meta AI since it launched so that people know that they are imagined with AI. And then Dall E3 is adding the C2PA, which is an open technical standard that allows publishers, companies and others to embed specific metadata into the media to verify the origin and to tell people that it was generated with AI. Camera manufacturers are also adopting this standard, as well as news organizations and others, so that as more images get thrown into the wild, be it through cameras, AI generation, or whatever else, this C2PA standard will help people understand what's real and what's not, theoretically. So if you're generating any Images with Dall E3 anytime soon, well, it's going to get this metadata. If I'm honest, I think there's probably going to be some easy workarounds for people to strip the metadata. So I don't know how helpful it's going to be, but it's a start and hopefully it's going to minimize some scams and fake images from spreading around to some degree. Also, in the past week, Neil Mohan, the CEO of YouTube, released a letter on the YouTube blog here about their 2024 priorities. Can you guess what priority number one was AI will empower human creativity. On their list of 2024 priorities, AI seems to be the number one priority for them. They've released a handful of various AI experiments into the YouTube platform. Things like Dream Screen and the Music AI Incubator and the ability to summarize comments with AI. But from the sounds of it, this is just the beginning of YouTube's AI journey. They have a lot more in store in the world of AI for YouTube. So as a YouTuber, I'm excited to see what they're bringing to the table. This week, Stability AI also released a new video model, Stable Video Diffusion 1.11. It seems to be a fairly slight improvement over the original Stable Video Diffusion, but if I'm honest, all of these AI video generators at the moment are still kind of looking the same to me. Now I've gotten some sneak peeks at some stuff in the works and I truly think models like these Stable Video Diffusion models are going to look fairly archaic in four to six months from now. Things are moving at a super rapid pace. The Stable Video Diffusion model is available on Hugging Face. We could see some examples of videos generated. Everything has that kind of slow mo feel to it still. And if you want you can download the files on Hugging Face. You just got to give them some details about yourself before it'll let you. But then you should be able to run this locally on your computer if you've got a GPU powerful enough. Me personally, I'm still going to places like Runway and Pika to get images generated because I feel like they're all kind of the same at this point. They all feel feel like they make very similar style videos. This is interesting. Hugging Face this week launched an open source AI assistant similar to Chat GPT's custom GPTs. So hugging face has their own sort of open source version of Chat GPT. Right now it's using Mixtral 8x7B, but there's various other models available that you can swap to whenever you want. Well, they've just added this little assistance option down in the bottom left and very, very very similar to custom GPTs. We've got some nice add ons that we can use directly inside of Hugging Chat. Things like an Italian grandma or a prisoner interrogation game or a bot that will mercilessly insult you. So if you like ChatGPT and the concept of custom GPTs, well maybe try Hugging Chat with an open source model along with one of these AI assistants which are also free to use and open source. And since I've said open source a lot in the last couple minutes. Let's talk about Smaug 72B. This is a brand new open source model. It's actually a fine tuned version of the Quinn 72B model. But inside these various benchmarking tests you can see that it actually performed better in almost every single area over Mistral Medium, Mistral Small, Gemini Pro and GPT 3.5. So these open source models like this Smog 72B, the Mistral of X Experts model and the recently leaked Mistral medium that was floating around, these things are getting really powerful. These open source models are getting better and better and better, some of them even catching up to GPT 3.5 Turbo Gemini Pro and in the case of the recently leaked version of Mistral Medium, that one was showing signs of being close to GPT4. Open source is really cranking along right now. This is a really interesting story that also came out this week. This was posted on Nat Friedman's Twitter account here. He said, ten months ago we launched the Vesuvius Challenge to solve the ancient problems of Herculaneum Papyri. Papyri, PI Papyri, Papyri. To solve the ancient problem of Herculaneum Papyri. I know, I butchered that. A library of scrolls that were flash fried by the eruption of Mount Vesuvius in 79 AD. Today we are overjoyed to announce that our crazy project has succeeded. After 2000 years, we can finally read the scrolls. Something that had been unread for 2000 years is now readable. And you might be asking, well, what does this have to do with AI? Well, in this article on Business Week from Ashley Vance called Can AI Unlock the Secrets of the Ancient World? Ashley talked about Nat Freeman. So Nat had a hunch that AI algorithms might be able to find patterns in the scroll images that humans had missed. Over the past few years, they began experimenting with AI. The team has scanned the scrolls with more powerful imaging machines, examined portions of the papyrus where ink was visible, and trained algorithms on what those patterns looked like. The hope was that I would start picking up on details that the human eye missed and could apply what it learned to more obfuscated scroll chunks. This approach proved fruitful, though it remained a battle of inches. They uncovered bits and pieces of the scrolls, but they were mostly unreadable. They put together a challenge and got a lot of other people involved and eventually it paid off and the AI systems had brought the past to life. Now I'll link up this article below because it's really well written, and it's a really good story of how AI was able to translate these ancient texts that were 2000 years old. But not all the AI news this week is good news. There was a massive heist using AI where a scammer got away with $25 million. Now, the way the scam worked was somebody got on a zoom call and there was a bunch of various deep faked members of the team, including a deep fake version of the chief financial officer and multiple employees of a company. It says the scammers were able to convincingly replicate the appearance and voices of targeted individuals using publicly available video and audio footage. Now, there's not a lot of details. They haven't announced what company got scammed, and they haven't exactly broken down what the scammers said to convince the company to pass along $25 million. But whatever happened, it worked. And these deepfakes and voice clones were were able to convincingly trick a company out of $25 million. And along the same vein, AI has been generating some pretty convincing driver's licenses lately. Now, if you've spent any time messing with crypto, if you want to sign up for an account like a Coinbase account, you have to do what's called KYC or know your customer. And during this process, you have to verify that you're a real person, that you're only creating one account. They want to see a picture of your driver's license along with a selfie to sort of verify that it's you. And as it turns out, you can now make these KYC documents and convince some of these sites that the ID that you're showing them is 100% legit. Meaning that the way a lot of sites verify you by using an ID and a picture of you is no longer going to be an effective method of verification. We're going to have to find some new ways to do this. Ironically, don't shoot me for saying this, but I do believe blockchain is probably the answer. There's probably some sort of blockchain tech that will be a much better way to verify who you are and that it really is you interacting with a website online. In fact, I think that's kind of what Sam Altman's World Coin was trying to accomplish. Whether that's the answer or not, I don't know. But photos of driver's licenses are no longer going to be something to verify that you are really who you say you are. They're too easy to fake now. But in good news, something that's going to minimize some of these scams is that the FCC officially banned robocalls with AI generated voices. Last week we talked about it, that it was something they'd been talking about doing. Well, this week they made it official. It is now 100% illegal to make robocalls with AI generated voices. I don't know how harsh the penalties are for committing this crime. And I also don't know how many of the bad actors that are actually trying to pull off scams and stuff are going to follow these laws. But it's a step in the right direction. And if you saw my video from earlier this week, you know that I got an Apple Vision Pro pretty much for the sole purpose of making video content with an Apple Vision Pro. I personally don't have a whole ton of use cases for it other than just playing with it and showing it off for video content. But here's something that I found really cool that's combining AI with something like the Apple Vision Pro. This is from Justin Means here. He says you can generate entire worlds with outtakes for Vision Pro. Immersive animated 360 worlds upscaled to 8K runs entirely local on Apple Silicon. All you need is a prompt and outtakes. AI will handle the rest. Coming soon to Apple Store. This could also be called the psychedelic simulator, right? It kind of gives you the impression that maybe you're on psychedelics or something and stuff is appearing and disappearing in front of you as you look around. But I'm in for it. I want to play with it. It looks really cool. And since we're on the topic of wearables for your face, this company, Brilliant Labs, is releasing these AI powered glasses. From the looks of this video, it looks like we're going to have augmented reality right on the glasses. They're going to be multimodal. You can use them with a prescription. It can do translation to other languages. You can look at the nutrition value of things. You can search the web with them. You can even generate images directly inside of these glasses. And if we take a peek at their website here, check this out. They're fully open source too. So the tech behind this and any code running on it, I'm assuming, is available for anybody to use. Let's go ahead and click on pre order now. I was just curious about the price. They're coming in at 349 bucks, which is actually comparable to the Meta Ray Bans. So taking a closer look at them here, it looks like the software that is going to be running on them are openly available models. Now it says visual analysis of the world around you, powered by OpenAI. That's not technically open source, but it'll be using an open API that most people have access to. It'll be able to translate the world around you, but that's powered by Whisper Searching. The web will be powered by Perplexity and you can hack and modify the frames because it's all open source. So these look pretty interesting. If these do everything they claim to do, these could be the best glasses on the market because they appear to be able to do what we can do with our Meta Ray Bans, but also what we can do with something like the xreal Air glasses. Xreal Air giving us the augmented reality capabilities, the Meta ray Bans giving us the large language model translation and sort of vision image recognition capabilities. You can do all of that in a standalone pair of glasses. These are going to be absolutely insane. And that's what I got for you today. Lots of AI news this week. It seems to be picking up steam. I'm expecting every week from here on out to have really cool, really big announcements that I can't wait to talk about. If you haven't already, check out Future Tools IO. This is the website where I curate all of the coolest AI tools I come across. I share all of the AI news I find. These videos, which I record on Thursdays, is just a little piece of all of the AI news that's rolling out. I have to pick and choose the most important things to to talk about because there's so much these days. But all the rest I do share on Future Tools under the News page. And I have a free weekly newsletter where I'll send you just the coolest tools I come across, just the most important news that you need to know for the week, and when you sign up, you get free access to the AI Income database, which is a database of various ways to make money using AI. You can find it all over at futuretools.IO a couple real quick announcements. I will be in Austin, Texas during South by Southwest as part of the Content Hacker live event on March 12th and 13th. If you're in Austin already or you're planning on coming to south by Southwest, you're going to want to be at this event. We're going to do a whole evening sort of fireside chat about AI, AI and marketing, the future of AI, all sorts of really cool AI stuff in person. We'll be hanging out, having drinks, eating together. It should be a really good time. I'll link up to where you can learn more about this event in the description. If you are thinking about going and you don't live in Austin, book soon because those hotels do book up fast. Also, I will be in person at Nvidia GTC this year up in San Jose, but they do have a free virtual event that you can attend. I'll make sure the details for this event are also linked up below. Nvidia is actually giving me an RTX 4080 Super GPU to give out to one of the subscribers who also registers for this free GTC event. So make sure you get registered. Maybe take a screenshot of your registration confirmation just in case I need to verify, but I'm going to be giving away a 4080 super to one person who registers for this completely free virtual event. And if you're going to be in person, hunt me down and say hi. And that's it. That's really all I got for you today. I hope you enjoyed this video. If you did and you want more news videos like this, you want AI tutorials, you want to stay in the loop with everything AI and future tech. Make sure you like this video and subscribe to this channel and that will make sure more videos like this one show up in your YouTube feed. Once again, I really really appreciate you and thank you so much to Tubebuddy for sponsoring this video. I will see you all in the next video. Bye bye.