The Most Insane Week of AI News in Months

Well, when it comes to AI news, October ended with a bang and November's off to a really good start. There's been a lot of cool stuff released just in the last few days and I don't want to waste your time, so let's get right into it, starting with a brand new update that came out from Midjourney. Midjourney now has what they call the Style Tuner feature. You could basically create your own style and then reuse that style over and over again. Now, if you pop into the Mid Journey discord, there's a complete tutorial here that walks you through how to use it, but I want to test it out myself real quick. If I jump into my Mid Journey room here and I type slash tune, I can now give it a prompt for a style that I want to recreate over and over again. For example, colorful glowing neon line art hit enter. And now it gives us some options here. Create a shareable style tuner based on your prompt to customize the style of your images. I can have it do it in either default mode or raw Mode, or select 16 styles, 32, 64 or 128 to go ahead and select 32 style directions and click Submit. It asks me if I'm sure because it's going to cost me 0.3 of my fast hours. Confirm and it does say it will take around two minutes for the tuner to be generated. You'll be notified when it's ready. And there we go. It gave me a link to my custom tuner here. So the next step is I open this in my Chrome browser and you can see now it's giving me all sorts of images and it wants me to select the images I like so that it will essentially tune this prompt. Now there's 32 sets of two, so there's actually 64 images to choose between. And I just select the ones that I think are good. So for example, I think these ones are cool here. I like the colors. I think these ones came out cool here. These are interesting. Either of these really match what I was going for, so I'm not going to select any of them. Same here. These ones are kind of cool. These ones are cool here. Keep on skimming to find the style that I really like. This one here is cool. Some of these are just like really far off from what I had in mind. Honestly, these here are pretty cool. I like the sort of glowing effect on them. These down here are cool. And you can see now that I've set my tuning, it gives me A code here. So I can actually copy this code and you can see that I add it to the end of one of my prompts, like this style and then my code. Now if I jump back into mid journey, I can do a wolf howling at the moon, and then I can put dash, dash style and then paste my style code in there that it gave me. And if I hit enter, it should follow the style that we just essentially tuned. Therehere we go. Four different images of wolves howling at the moon. They follow that style of my neon line art that I asked for originally. Let's do a mad scientist holding up a beaker in my custom style here. And this is what I got out of it. I could further tune this model by coming in here and selecting even more images and get it even closer to what I'm looking for. All right, so I did a little more tuning, selected a few more images here. Let's see how it changes things. So now I'm going to try a woman in the city with skyscrapers. And then this new style code. After I actually tweaked and picked some more images, it actually gave me a new code to use. So let's go ahead and try this new code. And here's what it generated. This time actually got a little bit further away than what I was hoping it would do. But this is my first time using this tune feature, so I definitely need to play with it some more. And once I really learn how to use it, I'll probably make a dedicated video just about how to use this cool new Mid generary feature. This week, Meta also started to give people access to their brand new EMU image generator. You can use it directly inside of Facebook messenger or you can use it inside of WhatsApp. You can get to it inside of WhatsApp by clicking on the little chat window up here and then selecting Meta AI. And then when you're inside of the Meta AI, you just type imagine just like you would with midjourney and say something like a woman in the city with skyscrapers. And pretty quickly we get our image of a woman in the city with skyscrapers. Now, it's not super realistic. Here's a wolf image I also generated. But it's another option for image generation. And as of right now, it's free. Free to use. Let's try. Imagine a photo of a man holding out his hands, showing them to the camera. Here's what that generated. It appears to be the right amount of fingers. There you go. Another potential free AI image generator that you can use it works in WhatsApp. It works in Facebook messenger, may work in Instagram Messenger. I haven't actually tried yet, but I know for sure it works in WhatsApp. Speaking of AI generated art, AI art got a little bit of a win this week. This week a U.S. district Court judge filed a decision in a copyright infringement class action law lawsuit that was brought against stability AI. Midjourney and DeviantArt. The AI image generator companies had filed a motion to dismiss the copyright infringement case this week. The judge largely granted it, writing that the complaint is defective in numerous respects. The biggest issue that he found was that two of the artists that made the complaint did not actually file any sort of copyright on their art with the copyright office. And then the other artist, Anderson, had only actually copyrighted 16 of the hundreds of works that they cited. Basically at the end of the day they concluded this because AI image generators reference art by many different artists when generating new imagery. Unless it is possible to prove that the resulting image referenced solely or primarily copyrighted art and is substantially similar to the original copyrighted work, it is likely not infringing on the original work. If this precedent holds up. Very few artists, unless they actually copyrighted their work and can prove that new images are using that copyrighted work and very similar to that copyrighted work, almost all of these should get thrown out, I would assume. Now there is a little bit of this court case still ongoing. The judge did allow one count and that was for the direct copyright infringement against stability AI for copying those 16 actually copyrighted works without authorization. So that piece of the lawsuit is still ongoing. But still, I believe Anderson has to prove that the new art generated is substantially similar and actually infringed upon those copyrights. So gonna be interesting to see how it plays out, but nonetheless a pretty decent sized win for AI art this week. Which bring me to the sponsor of Today's video, Wirestock IO. If you're not familiar with Wirestock, it is a single platform where you can upload your photos or your AI generated art or any sort of artwork. And the site will automatically title, write descriptions, add tags, mark it as AI generated if necessary, and submit it to all of the stock photography sites for you. And right now, many stock photography sites allow AI generated images. Sites like Imago, Adobe Stock 123RF, Dreams Time and Freepik all allow you to sell your AI generated generated art. Now, Wirestock is also really well known for their creative AI challenges, and this weekend they're doing a special 48 hour challenge. Simply create an image where you envision a world where cutting edge tools redefine daily life. From teleportation devices to mind bending contraptions. Let your imagination run wild as you create awe inspiring AI generated artwork that showcases the future of convenience. There's a grand prize of 50 bucks plus three runners up are going to get prizes of $20, $15 and and $5 for their outstanding AI generated art. So here's how it works. 1. Create AI generated artwork depicting innovative tools of the future that enhance the lives of tomorrow's inhabitants. Transport us into a world where convenience knows no bounds. 2. Participants can submit multiple AI generated images provided they have available submission slots. 3. Originality is key with this contest, so make sure the images that you submit are freshly created. Using preexisting artwork or any sort of plagiarism will result in disqualification. Or feel free to use all sorts of AI techniques. Whatever you need to do to bring your AI idea to life, feel free to get creative with the various AI art tools that are available. This challenge is going to go for 48 hours and the deadline is November 5th, so get your images created and in quickly over on Wirestock's website. Head over to Wirestock IO/challenges to get started. Here's some additional ideas and guidelines directly from Wirestock. Transport viewers to a world where television teleportation devices seamlessly whisk people from one place to another, revolutionizing travel. Imagine advanced household tools that streamline daily chores, making mundane tasks a thing of the past. Visualize wearable tech that enhances human abilities, blurring the lines between human and machine. Envision vehicles that defy conventional design, promising faster, safer and more efficient transport in the future. And here's a few little sample images they generated. So again, you'll be able to find the challenge over at Wirestock IO Challenges. And if you do decide to sign up and become a premium member of Wirestock, use the coupon code MATT20 for 20% off their premium plan. And thank you once again to Wirestock for sponsoring this video. Continuing on talking about AI art this week Google launched a generative AI tool for product imagery for us advertisers and merchants, so very similar to what we recently saw Amazon rollout. You can take your product images and add various backgrounds to them. Now, directly using Google's new AI powered Product Studio, you'd simply upload your product image with no background, describ the product, describe the scene, and generate multiple variations of your product. Last week I made a video about just mind bending AI tools that are becoming available and A big topic of that specific video was how cool some of the text to 3D stuff that's coming out is. Then when I did a live stream earlier this week, while I was on the live stream, Luma Labs actually messaged me to tell me that they just released their new text to 3D model called Genie and this one is available directly inside of their discord. Jump into the Luma discord, come into one of the these Genie rooms, type slash genie and then enter the prompt of what you want to generate. I'm going to give it the prompt a fighter jet and literally less than a minute later I have some 3D examples of fighter jets and they all look pretty dang good. I think either this top left one or this bottom left one nails it the best. But if I click on number three here, it will jump me over to the Luma Labs website and you can see I've got a 3D object that I can click and drag around and zoom in on. And then if I go to click to download it, it downloads it as a GLB file, which is a 3D object file that I can then pull into Blender and clean up if I want. I've said it on my live stream, I said it on my video about mind bending cool AI tech that's in the works. But man, text to 3D object has come so far just in the last few weeks. It is honestly mind blowing. Here was the other one that I thought looked pretty good as well. I could zoom in, do whatever I want with this. I think it would even be fun to pull these into a slicer software and try to 3D print some of these, maybe after cleaning them up a little bit. And While speaking of 3D stability, AI this Week announced that their latest tool is going to use AI to generate 3D models. They think that 3D model creation tools could be the next big thing in generative AI. And I tend to agree. We're seeing just crazy advancements now. Right now their 3D model is only available in private preview for select customers who reach out to Stability via the company's contact form. So, you know, I'll be reaching out and asking if I can get access and make a future video on it. But apparently you can go to their contact form form and give it a shot, see if they'll hook you up. Here's an animated image that they actually released showing all of these 3D models dropping into a bowl. You could see there's a whole bunch of them. According to this tweet thread Right here, they said this scene and all these models were created from text prompts and generated in a few hours. So we know it'll be fast. And you can actually see that this video here is a lot more models actually dropping into the bowl than this animated GIF that we were watching here. If they generated all of these models in a few hours, well, it seems like it's going to be pretty polished and pretty quick generation. Super excited to get my hands on that one. Continuing on with 3D, Blockade Labs posted this tweet this week about how they had a new AI breakthrough. They've auto nerfed 2D 360 degree images generated with their Skybox AI into 3D explorable worlds to include free roam and dynamic lighting. So they actually generated a light box inside of the Blockade labs, you know, 360 degree generator, and then managed to use Nerf technology to make it so that you can move around in that scene and it would change that scene to 3D space. That, to me, is pretty crazy. Just check out this demo. As they move around underwater in 3D space. Look at the water reflection as they look up through it. This is pretty crazy. This was all done with essentially a text prompt to a 360 degree image and then taking that 360 degree image and somehow turning it into a Nerf. All right, let's talk text to video for a minute. This week, Runway actually updated their Gen 2 model to make the videos that it generates so much more realistic. I came across this tweet from Max here. This is massive update. Runway has launched something magical. I feel like this is a pivotal moment in Generative AI. I didn't see any announcement about it. Just a massive but silent update from Runway. The quality went up immensely. Take a look for yourself. And here's some of that AI generated video that was done directly with the newer model of Runway. And it's looking really, really impress. I'm pretty sure you haven't seen text to video that looks this good yet, because I know I haven't. Now, they mentioned that they silently made this announcement. This was published at 3am on November 2, and eventually Runway at 6am on November 2 made their announcement on Twitter X. We have released an update for both text to video and image to video generation, with Gen 2 bringing major improvements to both the fidelity and consistency of video results. And here is some even more example examples of what this text to video is now capable of. Now, to use this, simply go over to Runway over@runwayml.com log into your account and then right at the top here they have Bring your image to life with Gen 2. Start with an image or start with a text. Let's quickly try both. I start with text. I'll do a gray wolf walking through the woods with hints of snow on the ground. And here's the video it generated. Now it just kind of shows the wolf looking around. I sort of wish it made the wolf walking, but it's a pretty dang realistic looking wolf. Let's go ahead and upload an image. Now I'm curious what happens if I just upload a headshot of myself? Will it try to animate my headshot? And here's what it did with my face. It essentially morphed me into a different person, but still pretty dang realistic. Managed to keep the shirt looking the same. Still really, really cool. Gen 2's results are a heck of a lot better than what they used to be. You can always extend by four seconds. You can do that up to three additional times, allowing you to have 16 second generations. And they do have the option to combine an image plus description. Also in the text to video realm, if you've ever used Kyber to turn your existing videos into cool, really stylized videos, I love Kyber. They just released mobile apps this week on both iOS and Android, so now you can shoot video directly on your phone and use the app to start actually creating a new version of the video like right away. Now I haven't personally used this app yet, but I am definitely going to be downloading it and playing around with it this weekend and I will report back at least on Twitter. All right, shifting into large language model news, this week OpenAI started rolling out their new all in one ChatGPT model. Personally, I'm still a little bit salty because I don't have it in my account yet and I keep on seeing people on Twitter post some of the cool stuff they're doing with it. But essentially what this new update's going to do is if I hover over GPT4. Here you can see Browse with Bing Advanced Data Analysis plugins and Dall E3. It rolls them all into just one so you don't have to switch between them anymore. You just prompt to ChatGPT and it will use the best model based on your prompt. So you'll be able to use the Vision model where you upload an image and then tell it to create another image similar to that. And it will use Dall E to generate more images similar to that. You can upload PDFs and have it summarize the PDFs for you. And rumor has it the new all in One model of ChatG is actually using the 32,000 context window, meaning that you can add a lot more text and get a lot more text back out of Chat GPT once you have this new updated version. Now, I haven't 100% confirmed that yet, but some people who have gotten access have already skimmed the code and claim that it is using the 32,000 context windows, so fingers crossed that that's true. This coming week OpenAI is actually having their first ever development day, and I'm expecting to get some big news out of that day. So next week will probably be a big week for OpenAI, and we'll probably also learn a little bit more about this new all in One model that they're rolling out for us. At the end of last week it was also announced that Google Bard can now respond in real time instead of you having to wait for the response. Now, this isn't huge news, but essentially what it means is if I say create a recipe for Key lime pie for me and hit enter, you'll actually see it start to write it out in real time instead of just a blank square here until it's fully generated. So if you start to see it generate and you're not happy with the generations, you can actually stop it mid generation and have it start over. You can see I even got a notification window as this was typing responses will now show in real time. You can change this in the settings at any time, so if you go to your settings you can have it response once complete or have it respond in real time if you're a fan of using the Brave browser. Brave is actually working on their own AI chatbot that they're calling leo, which I think is really interesting because if you ever look at Feedly, they have their own AI that they call Feedly leo. So I wonder if there's going to be some butting heads in the future over that name, but we'll see. It doesn't sound like the Brave AI is going to do a whole lot more than what you'd get out of something like Bing or ChatGPT. However, they're claiming that their version is going to have unparalleled privacy compared to some other AI chatbot services. And if you use the Brave browser and you Update to version 1.6 of that browser, you should now have this LEO directly in your browser. I don't personally have Brave installed, but I sort of anticipate that it's going to work similar to the way Bing or ChatGPT does. It's just not sharing any of the data or using any of that data for training purposes. This week Instagram also had a little leak around a new large language model for their customizable AI friend. Here's some leaked screenshots of what this Create an AI inside of Instagram looks like. You can see on this left window here it says make an AI you can get started once you start creating it. You can select the gender, female, male, non binary, age, young adult or elderly. You can pick an ethnicity for it. You can select personalities for it. Reserved, enthusiastic, creative, witty, pragmatic, empowering. It then looks like you can select interests for the AI. You know, DIY art, design, celebrities, gaming, entertainment, music, etc. You can then select an avatar for that AI, name the AI, and then start chatting with the AI. Now this hasn't been 100% confirmed and of course Instagram declined to comment on the matter, but this seems to be coming from a pretty reputable source. So I think it's pretty safe to say that we're not far off from having an AI builder inside of Instagram. There's another AI model on the scene called Find P H I N D and according to this release here, it actually beats GPT4 at coding and has similar speeds to GPT3.5. It's also got a 16,000 context window which is larger than the context window that we actually get inside of Chat GPT right now. And this Find tool is available right now. I'll link this blog post in the description of this video along with all of the other content that I've mentioned in this video. But if you come to this post and you click on Try the Find model here, it opens up a chatbot and you can see right here on the right you can either use the Find model or you can use GPT4 and it actually lets you compare the results to find and GPT4. It only lets you use GPT4 10 times total. And the Find model has unlimited uses, but I'm pretty sure it's here so that you can compare the results of the output you're getting getting. But if you've been currently using chat GPT with gpt4 or writing code, maybe head over to find.com and see how this does with your coding as well. I'm in the process of working on some games using AI. I think I'm going to give this Find a try and see how it compares when actually trying to create my game. This week, the AI chatbot PO from Quora did something really interesting. They're actually allowing people who help them create the bots to monetize the bots that they created, essentially creating an AI chatbot creator economy. It says the program, which went live last week, is among the first of its kind to monetarily reward the efforts of those building AI bots. There are two ways for bot creators to generate income. If a bot that you create leads a user to subscribe to po, the company will share a cut of the revenue back with the bot's creator immediately. Another method involves bot creators setting a per message fee, and Quora will pay on every message. At the moment, this creator monetization program is only open to US users, and it pays up to $20 per user who subscribes to PO thanks to a bot that you created. Now, I personally haven't created any bots yet myself, but here's a quick demo that they showed off of how you would go about it. You can edit your own bot with bot description, your style prompts, intro messages, bot access, the custom message limit, and once you've customized it, you can then copy that link, share your bot anywhere. And then of course, if somebody decides to sign up to PO because they want to use your bot, you that $20 cut, assuming you're in the US but it does sound like it's going to roll out other places. And in the last bit of large language model news this week, Google commits $2 billion in funding to AI startup Anthropic. Now, Google was already an investor in Anthropic. I believe they originally invested half a billion. But this deal is interesting because obviously Google is working on Gemini with DeepMind. They have their own Palm 2 model. So Google has other AI models. But I think they're kind of hedging their bet a little bit because right now, for the most part, it seems that Claude is actually a better bot than what Google currently has available. But it also sounds like Google wants a piece of the pie. Anthropic is a big competitor to ChatGPT right now, so they'll have their own models and they'll have Anthropic in the same way that Microsoft invested in OpenAI but also invested in Meta's llama. Both these big monster companies are essentially hedging their bets by investing in multiple different different AI chatbot companies. Speaking of mega companies, the mega ist of the companies Apple had an event this week called Scary Fast. During this event, they introduced their new M3 series of chips. They introduced their new MacBook Pros and they introduced their new imacs, but really these M3 chips were the star of the show because the imacs and the MacBooks that are all rolling out are rolling out with these M3 chips. So they're essentially the same Apple computers you're used to just now with M3 chips in them, which are just these insane processors. Believe it or not, during this keynote they actually mentioned AI. I counted twice. They also talked about neural networks, they talked about machine learning. All of the AI buzzwords were actually in this keynote where Apple seems to try to avoid a lot of those AI buzzwords in most of their keynotes, but it seemed like with this one they finally started embracing it, talking about how their new chips and their new technology are good for training and leveraging AI technology. It sounds like their Apple silicon chips are being designed with AI in mind. Now this week President Biden and the White House issued an executive order on safe, secure and trustworthy artificial Intelligence. This is a really long statement here. Ben Tossel over at Ben's Bytes did a nice quick rundown which I will also share in the description below. Any model trained with roughly 28 million H100 hours, which would cost around $50 million USD, would be limited. If you have roughly 50,000 H1 hundreds, you're also going to have some limitations. The order also directs different agencies to develop new standards for AI safety testing and monitoring, securing critical infrastructure with AI, preventing fraud and ensuring fairness in criminal justice. AI. My buddy Bilavel shared a TLDR on his Twitter account about all of these various government agencies and acronyms and what they are each individually supposed to do. The NIST will set standards for AI safety testings and DHS will establish an AI Safety Board to review critical infrastructure. The DOJ and Civil Rights offices will give guidance on avoiding algorithmic discrimination. HHS will support privacy preserving techniques and evaluate commercial data procurement. DOL will develop principles to protect workers from AI harms. NSF will fund research on privacy preserving technologies. State Department and Commerce Department will collaborate internationally on AI safety safety. But I also retweeted this post from Bendu Ready this week, which does a pretty good job of sharing why this executive order is probably not going to have as big of an impact as most people believe. One of the things that was suggested is that all AI generated content has to be watermarked, she says. Seriously, we may as well kill vision AI if we actually enforce that. Are enterprises allowed to use AI to Generate images and use them in their marketing. This is kind of one of those things where, where if you require every AI image to be watermarked, well, does that mean we should be led to believe that if it has a watermark it's always generated by AI and if it doesn't have a watermark then someone didn't figure out how to circumvent the system? I mean it doesn't really seem feasible to enforce that or even keep up with it because people are going to figure out how to watermark non AI images and people are going to figure out how to remove watermarks on AI generated images. It's just inevitable. Some agency is apparently going to develop tests that the AI models have to pass. How will this get enforced? Does every fine tuned version of any open source model have to pass these tests? What happens if a model is dropped on hugging face by a non US corporation? Again, how do you enforce that stuff? There's a bunch of privacy protecting language. All this does is cripple AI models compared to search where the same privacy isn't being enforced. Google search will recognize a face for me. GPT or Vision won't. It's not clear why my photo that is in the public domain can't be used by AI. We already have privacy protecting laws. These should be sufficient. She goes on to say net net. This order won't make any difference in the near term. Could be harmful in the long run to companies who do develop these foundation models, especially vision models. Continuing on with the theme of AI safety, there was the AI Safety Summit in England this week where China, US and the EU all agreed to work together. This summit included members from China, the European Commission, India, United Arab Emir and Nigeria. Basically all these countries got together saying that we all understand the risks that AI could create. So let's make sure we're all on the same page about creating it safely and ethically and trying to also minimize various biases together in these AIs. Once again Ben Tossel from Ben's Bytes put together a nice little breakdown of what you need to know. Again shared in the description. Different countries apparently have different areas of focus. So. So the US is going to get companies to make voluntary commitments and adhere to reporting requirements, launching an AI Safety institute of its own under the Department of Commerce. China is working on equal rights for every country to develop and use AI. Global cooperation to share AI knowledge with the public on open source terms. EU is working on innovation, opening up EU supercomputers to train models for free of charge. They're working on guardrails and governance. India AI is, is working on AI as an enabler of mass digital adoption and so on. There's various countries involved all committing to work on different aspects of these challenges. This week the Beatles dropped their last Beatles song. It's called now and Then. You can actually find it on YouTube. The song was actually created with the assistance of machine learning. They took some old tapes of John Lennon, were able to use machine learning to separate his audio from the rest of the music and then clean up the audio and then produce a new track with his vocals. Pretty cool little thing that we get to witness in that we get to the last ever Beatles song with John Lennon actually singing on it. In a attempt to be part of the AI hype and buzz, Doritos devised an AI powered crunch cancellation technology for gamers where you can eat your Doritos chips while on a game and it will mute out the sound of the chips crunching. I guess where in reality what you really need is like something to not get Doritos dust on your fingers because who wants to play games touch their keyboard and mouse with Dorito hands. This is really cool. AI can diagnose type 2 diabetes in 10 seconds from your voice they trained an AI model to diagnose type 2 diabetes in patients after 6 to 10 seconds of listening to their voice, they trained it to recognize 14 vocal differences in the voice of someone with type 2 diabetes compared to someone without diabetes. The auditory features that the AI focused on include slight changes in pitch and intensity which human ears cannot distinguish. This was paired with basic health data gathered by the researchers such as age, sex, height and and finally, in the last bit of fun news for the week, AI was named the word of the Year by Collins Dictionary. Apparently use of the term has quadrupled this year alone. They said that we know that AI has been a big focus this year in the way that it has developed and has quickly become as ubiquitous and embedded in our lives as email streaming or any other once futuristic now everyday technology. So that's pretty cool. And if you want to find some other pretty cool tools, check out featuretools IO that was a ridiculous transition. This is where I cur create all of the coolest AI tools that I come across on a daily basis. I keep up to date on the news every day and I have a free newsletter that you can join and I will send you the latest AI news and the coolest AI tools directly to your email inbox. You can find it all over at FutureTools IO if you want to stay in the loop with the AI news. You want AI tutorials, you want to hear about the latest cool AI research, Make sure you like this video and subscribe to this YouTube channel channel and I'll make sure more stuff like this shows up directly in your YouTube feed. Thank you so much for tuning in. I really, really appreciate you. I hope you enjoyed this video. I love nerding out about this stuff. I hope you enjoyed Nerd now with me. That's all the news I got for you this week and I'm recording this on Thursday. So there might be new news on Friday that I didn't get a chance to talk about, but if there was, I'll make sure it's in next week's video. Thank you so much for tuning in. And once again, thank you to Wirestock for sponsoring this video. I really appreciate you guys and everybody who spends the time to watch these videos and makes it this far to the video. You rock. I love you guys. I'll see you in the next video. Bye Bye.