AI News: GPT-4-Level Open Source, New Image Models, Neuralink (And More)

AI news has actually been kind of slow so far. For the last couple weeks we had CES where a whole bunch of big announcements were made and a whole bunch of cool products were revealed and then it was just kind of quiet. There wasn't a lot to talk about. However, things are really starting to pick up steam again, and as we roll into February here, there's really been some kind of exciting new developments and tools that we get to play with now. So let's jump right in.

Starting with the fact that we can now generate images directly inside of Bard. Bard is finally catching up with ChatGPT, at least in the ability to generate images directly with inside the chat platform. If we take a peek inside of the bard.google.com updates page here you can see on February 1st they rolled out the update where we can now create images with Bard. Now it is important to keep in mind that it's not yet available in all countries. According to According to this support docs, it says image generation in Bard is available in most countries except in the European Economic Area, Switzerland and the uk, and right now it only works with English prompts. But if you want to start generating images within Bard, it's simple. Just head over to bard.google.com, you'll get the chatbot that you're used to directly in Bard and you just tell it what you want from it. For example, generate an image of Mario kite surfing. This time it generated just one image. Usually it generates two images. Not sure why it only generated one, but it does give you this button to generate more, and after a few seconds it'll generate more. Once again, it only generated one instead of two more, but those are clearly Mario Kite surfing. Now one thing that I've noticed when using Bard, it denies a lot of prompts, like prompts that it doesn't seem like it should be denying. When I posted about it on X today, that was pretty much the number one complaint people had about it. So Siris here gave an example where he says I want images of a brown cow drinking chocolate milk in a field. And it said, unfortunately I can't generate images that are physically impossible or misleading. Like a cow drinking chocolate milk. Cows are lactose intolerant, meaning they can't digest milk properly. While the image is a fun thought experiment, it wouldn't be accurate to reality. Like but if I try to ask it to generate an image of any sort of real person, like generate an image of Tom Hanks, it says I can't generate images of that. Try asking me to generate images of something else. Yet when I ask it to generate something that's clearly trademarked IP like Sonic the Hedgehog, it has no problem with that very hit or miss on what it will actually let you generate. There's been some really random prompts that it should have no problem generating. Nothing offensive about them and it just says it can't. For some reason my buddy Mattvid Pro actually made a video all about what Imagen 2 is capable of. And he was getting a lot of the same results in his video which where he was asking to create images with a human with eight arms and things like that. And it was saying I can't generate that. That's outside of our policies or whatever. I also learned from this video from Matt Vid Pro that you could generate images directly inside of Google's Test Kitchen inside of their tool called Image fx. Google's AI test kitchen is where they release a lot of their AI projects early and they allow testers to come in and play around with them before they are pushed out mass to the whole world. And if I go to Test Kitchen and scroll down, I can see that we have Image Effects here. And this is a purpose built image generator. Obviously Bard is a chatbot with Imagen built into it. Image Effects appears to be a user interface solely for the purpose of generating images. With Imagen 2, I'd like to say it gives you a little bit more control than trying to generate the images directly in Bard, but really the only additional options it gives you are the ability to set a seed. Down here I'm feeling lucky button which will generate a random image for you such as a turtle made of potato chips. Actually pretty good. And it's got these extra little like prompt helpers down here so I can add things like handmade 35 millimeter film. So it gives some suggestions to improve the prompt and under each suggestion it also gives you a little dropdown. So instead of made of you can try sculpted out of painted on, drawn on instead of handmade digital painted sculpted instead of 35 millimeter film. We've got some other options here, but other than letting you set the seed and giving you some sort of prompt helpers, there's not really much more control. But it does generate four images at a time instead of two without you needing to press the generate more button. And if you're curious what this prompt looks like. A turtle made of potato chips. Handmade 35 millimeter film. There you go. I actually thought the original was better. One thing the other Matt figured out in his video was that Imogen is really good at sort of photorealistic Images, almost midjourney version 6 level, but not quite as creative and also not quite as prompt adherent as Dall E3 inside of ChatGPT and to sort of pile on the negatives. He also found that a lot of random prompts that should have worked. It told him it couldn't generate for some reason. So I don't know, play around with it yourself, have some fun with it, see what you can generate. It's free to use right now. What's not free to use is Midjourney and Midjourney just rolled out a brand new AI image generation model. If you're into generating cartoon and sort of anime images, they just rolled out the Niji V6 model which generates these sort of anime style images. If you have a Midjourney account and you want to play around with this, jump into midjourney, type slash settings. Click on the model dropdown here and we can see Niji Model 6 Alpha select that dismiss the message and now now when I generate a prompt it will prompt it in this NI mode I could prompt something like a human cyborg battling a human ninja and you can see it added dash dash NI6 at the end. And alternatively if you don't want to change it in the settings, you just want to generate one image. You can also add this dash dash nii6 to the end of your prompt to force it into this NI mode. And here's what we get when we generate a cyborg battling a ninja. While we're on the topic of AI image generation, Shopify is rolling out the magic AI image editor and here's what that does using Shopify's own words.

Starting today we are rolling out new Shopify Magic features in the Media Editor that use generative AI to make product image editing and enhancement easier than ever. All right here in the admin. With these new AI powered background editing features you you can select from common style presets for an instant boost of professionalism or reimagine the scene altogether with a short description of what you'd like to see or instantly match the background of your existing photo shoots to make your storefront feel more on brand and cohesive.

For today's video, I partnered with HubSpot to tell you about their free campaign assistant powered by AI. So check this out. If I click on Get Started it brings me inside my campaign assistant here. If I click on Create new Asset, I have the option to create a landing page Google Search Ad marketing, email, Facebook ad or LinkedIn ad. Let's go ahead and create a Google search ad over on the left side here. It asked me a few questions to get an idea of what my business is all about. So I filled in some details about my website. A website that curates all the latest tools in the artificial intelligence space. I added three bullets of what I want my audience to know about my business. We review all the latest tools in the AI space. We sort through the bad and sketchy tools and only surface the good stuff. And we try to work out discounts and deals with the creators of the tools to save you money. I can add some keywords here. I added a handful. And then what? What do you want your audience to do? I'll use the call to Action Browse for free today. Under writing style we have all sorts of options. Witty, creative, humorous, friendly, positive, et cetera. I like to go with creative, humorous and friendly because that's how I like to think of my brand. And let's generate content after just a few seconds, it generated three Google search ads for me. Discover all the latest AI tools. Find and share unique AI tools at future tools. Save money with discounts from creators. Browse for free today. Now I can simply copy these ads and quickly paste them into a notepad here so that I could use them in my Google Ads account. Now if I go to, let's say, create a marketing email, it actually already remembered everything that I've already entered about my business. So all I have to do is click Generate content and a few seconds later I have a marketing email that I can copy and paste and send to my email list audience. And it works the same way for LinkedIn ads, Facebook ads, and even landing pages. All of my details are pre filled in. I just generate content. It generates the copy for my landing page. I can click Create Free Landing page and with a single click I have a landing page with my copy already loaded in. Now all I got to do is start fine tuning it and swapping out some images. HubSpot's free AI campaign assistant really speeds up the process of creating your marketing assets for your business and makes it super easy to generate everything you need to get eyeballs on your product. Again, Campaign Assistant is totally free to use. I will make sure the link is available in the description for you so you can try it out today. And thank you so much to HubSpot for sponsoring this video. Now this next thing isn't something we actually have access to yet, but I thought it was pretty cool. Microsoft released some research called Stroke NUA or Stroke N U W A. And basically it's a new way of sort of prompting images, more like prompting drawings, I guess, where it turns certain line shapes into vectors in a vector database and then you prompt it using a large language model. It's sort of like generating AI images in the same way you generate text with something like GPT4 in chat GPT instead of a diffusion model or a GAN Generative Adversarial network model, like what the other image generators use. Again, this is still early research. We don't really have access to play around with this yet. But there is some cool sort of examples of the type of images this will be able to generate. And yes, they kind of look like children's drawings right now, but we're basically asking a large language model that's similar to like a ChatGPT type model to actually draw pictures for us and not generate them through the standard diffusion models. It's a completely new approach to generating these images and I'm really excited to see how this plays out and how this improves in the future. Here's some more interesting research that was recently made available on Hugging Face for us to actually play with. It's called Image to Sound Effects or Image to sfx. You upload an image, submit the image and it will generate a little sound clip to go along with that image.

Foreign.

Was generated with nothing more than the image that was uploaded. This week, Chat GPT rolled out a brand new feature where you can actually call upon GPTs in your chat and have multiple GPTs all inside of a single chat. I actually did an entire video breakdown of this new feature. You can find it by looking for this video right here. But basically when I'm inside of the Chat GPT user interface, I can type at and it will bring up a list of all of the various GPTs I've used recently. I can select one, prompt it with something like what's the most recent paper about AI? It gives me a response to that, and without ever leaving this chat window, I can call upon another GPT. For example, the diagrams show me GPT and tell it to create a diagram based on the last response. Just like that, it generated a diagram for me all in the same flow and the same chat as the first prompt that I gave to Consensus. So you can actually start to tie and blend various GPTs together to get really cool final outputs and do some really interesting things inside of Chat GPT. But again, I have a full breakdown video of it. It looks like that While we're on the topic of large language models, something really interesting happened this week. A new open source large language model appeared on Hugging Face for people to use and it was giving outputs that were near GPT4 level outputs. The model was called Miku M I Q U. According to this post from N8 Programs here, it gets an 83.5 on the EQ benchmark, surpassing every other large language model in the world except for GPT4. It even beats Mistral Medium which is supposedly Mistral's best model right now. Now when this model appeared on Hugging Face, there was a lot of speculation that this was actually a leaked version of the Mistral media model. That's kind of a tongue twister and that possibly Mistral themselves leaked it. And when people started doing tests with it comparing it to the Mistral Medium model that's available on Perplexity, it was pretty much giving the exact same outputs which all but confirmed that it was a leaked version of this Mistral medium. Well this week the Mistral CEO then confirmed that the leak of this open source AI model is actually one of their Mistral models. In this article it says Today it appears we finally have confirmation. Mistral co founder and CEO Arthur Mensch took to X to clarify an over enthusiastic employee of one of our Early Access customers leaked a quantized and watermarked version of an old model we trained and distributed quite openly. We retrained this model from llama2 the minute we got access to our entire cluster. The pre training finished on the day of Mistral 7B release. We've made good progress since basically all of this to say that Mistral took a version of llama 2, sort of beefed it up, made it better, started getting near GPT4 results and then somebody over on the Mistral team leaked it onto Hugging Face and now it's accessible for people to actually play around with. And according to the CEO, it's actually one that they trained a while ago and they actually have better models now. So Mistral probably has some game changing AI in the works that could be pretty exciting. While we're on the topic of open source large language models, this week Meta released their Code Llama 70B model and according to one of Meta's own posts here, Code Llama 70B instruct achieves a 67.8 on human eval, making it one of the highest performing open models available today. And code llama70B is the most performant base for fine tuning code generation models. And if you don't know what that means. Don't worry, I didn't really either. But here's a post from Philip Schmid here on X, basically explaining it as code llama70B was able to reach initial GPT performance. So when GPT4 first came out it was pretty dang good at code. Quite a leap from GPT 3.5. We do now have GPT 4 Turbo, which is probably a little bit better at writing code. But this code llama7DB is about as good as that initial GPT4 and it's open source, anybody can use it, and anybody can fine tune it to their own coding needs and commercial uses allowed. So you can build your own software with it and resell it if you want. There's also been a lot of talk lately around Apple and them getting deeper and deeper into the AI game. Now this article here on QZ.com is actually from September of 2023, so this may actually have changed a little bit since this article came out. But if we look at this graph, Apple has actually acquired more AI companies than pretty much any other company we can see. Since 2017, Apple had acquired 21 AI companies versus 19 from Accenture, 12 from Microsoft, 11 from Meta, and then Alphabet's all the way down here at 8. So although Apple doesn't talk about AI quite as much as the rest of these companies, they're quietly accumulating all of these AI companies. And this week on February 1st, Tim Cook confirmed that Apple's generative AI features are coming later this year, saying that they're putting tremendous time and effort into integrating AI into its software platforms. Now Apple sort of comically never mentions AI in any of their keynote presentations, but in a recent earnings call with Tim Cook, he apparently mentioned generative AI several times, but never really actually got too specific. So we don't actually know what that means. We don't really know what sort of generative AI is going to be added into Apple products, but the biggest speculation right now is that we're going to get a new version of Siri, most likely announced at this year's WWDC event. But Tim Cook was very hush hush on exactly what AI is rolling out, so we're just going to have to wait and see. And since we're on the topic of Apple, this week is a fairly historic week for Apple. They are launching a brand new product, a brand new gamble for their company with the Apple Vision Pro. It seemed to get into a lot of other youtubers hands, not mine, but it's going to be available for general consumers starting this week. In fact, the day that this video goes live is the day that most people who pre ordered the Apple Vision Pro is going to get access to theirs. I'm a huge sucker for new tech, so I ordered one and I'm expecting to have mine hopefully by the time you're seeing this video. And along with the Apple Vision Pro release, Apple did announce that more than 600 new apps that are purpose built for the Apple Vision Pro are going to be available at launch, so they should be available when we get access to our Apple Vision Pros. I'm not going to go into too much detail on the Apple Vision Pro in this video, since by the time I make another video I will have had access to one. I should be able to dive a little bit deeper into what I know about Apple Vision Pro at that point. This week we also saw a new mobile browser called Arc Search. In fact, if you watch my other video all about really cool AI apps available on mobile phones, which looks like this video right here, you've already seen me play around with Arc Search, it's basically a browser where you can enter any prompt you want. For example, I type San Diego Padres, I click Browse for Me and then it generates a pre built webpage for me with everything I would want to know about the Padres, including images of the stadium, details about the team, their home stadium, their achievements, the top search results, details about the franchise history, details about the current season, details about notable players, upcoming games. It basically creates a single webpage with everything you'd want to know all on that web page. So you don't need to click around and browse different web pages to find what you're looking for. If I'm honest, I think it's a really cool app. I think this is sort of what end users want. I don't think people want to have to dig through a whole bunch of websites to find the exact thing that they're looking for. It's nice to just type in a topic and just get all of the details about that topic presented right in front of you. However, it does sort of present a problem to content creators because if I'm creating content and tools like this are just surfacing the information people are looking for, there's no incentive to click over to my website, which means I might miss out on ad revenue and the ability to build my mailing list and the ability to offer products on my website. It kind of disincentivizes content creators. While I like the concept and from a user standpoint, it does seem to make my life easier, but from a content creator standpoint, I have some concerns, but I'm curious what you think. Let me know in the comments. This week the New York Times announced that they're building a team to explore AI in the newsroom. If you remember, the New York Times is one of the companies suing OpenAI right now for scraping content from the New York Times, but they're putting together their own internal AI team so that they can explore generative AI and what it could possibly be used for within the New York Times. They do, however, say that journalists will still write, edit and report the news. If I had to guess, it sounds to me like the New York Times wants to build a New York Times bot. So instead of you prompting to find news inside of ChatGPT, if you want to find information specifically from the New York Times, you can go and ask the New York Times bot and it will be trained on all the various New York Times articles and respond based on the information that's solely within the New York Times database. That's sort of my guess of what's going on here. Here's something that I don't think anybody's going to complain about, except for maybe scammers and spammers, but the FCC is moving to outlaw AI generated robocalls. Now they're doing this to try to prevent and minimize scams and spam calls and things like the fake Biden calls. But if I'm being honest here, bad actors are going to be bad actors and even if it's illegal, they're just going to do it anyway. So I don't know if outlawing it is really going to have an impact on the people that are using this technology to scam others, but pretty soon it's going to be on the books that it's illegal to do that, although I already think most people see it as pretty highly unethical. This week we also got a robot that was trained to read Braille at twice the speed of humans. The research team from the University of Cambridge used machine learning algorithms to teach a robotic sensor to quickly slide over lines of Braille text. The robot was able to read the braille at 315 words per minute at close to 90% accuracy. Now they say that the robot braille reader was not developed as an assistive technology and it was really designed as more of this like test to find out how good sensors are, almost like a benchmarking system to test future systems to see how well they can read Braille. However, this article does go on to say that they are thinking about putting this technology in inside of the fingertips of humanoid robots so that they can read braille as well as into things like prosthetics. But it does sound like the initial use case for this is more for robots than for humans. In the last week, we learned about a new AI model called Morpheus One, which claims to induce lucid dreaming. It's basically this headband that they call the halo. And the halo sends sound waves, or ultrasound holograms into the brain to connect with the current brain state and combined can put the mind into a lucid state. Now, if you're not familiar with lucid dreaming, Lucid dreams are a type of dream where the dreamer becomes aware that they're actually asleep. So when you're in the dream, you realize that you are asleep and in the dream, effectively giving you control over the dream. The company also says they're going to start beta testing this thing in spring of this year, 2024. This is one of those technologies that I think is really cool, but I have a hard time sort of comprehending it working, if I'm honest. Like, I would love to see it, but for some reason I still feel very skeptical about it. Hopefully I can be an early beta tester on something like this because I would love to try it and be proven wrong and be able to have more control over my dreams, because that just seems really cool. And finally this week, in some pretty major news, neuralink from Elon Musk was implanted in the first human patient. Now, the neuralink itself is this little circular chip that's about the size of a quarter. It's actually designed to be implanted in your skull and actually help your brain interface with a computer. Now, the first use cases for this product are for quadriplegics to help them regain some sort of functionalities that they lost. But I feel like it's only a matter of time before we start getting that, like, matrix scenario where you could just download how to do Kung Fu into your brain. Okay, so there's one last thing that I want to do in this video today. Inside of the Future Tools Discord, we ran an AI video generation contest for the new year, and January 31st was the deadline to get in your video submission for the chance to win 500 bucks. Inside of the Discord, we got tons of amazing submissions, and John, my creative director, compiled the best ones here into a little video reel so that we can watch them. I have not watched these yet. I have no idea what these videos are going to look like, but I thought I would watch them for the first time on this video. And then John Zeno and I inside of the Future Tools Discord. We're going to narrow it down to, like, our three favorite videos and then we're going to put it to a vote to see who wins the $500. So if you're not in the future Tools Discord yet, jump in there. If you are in the future Tools Discord and you're watching this right now, we'd love your say so. Make sure you get into the Discord when we announce our top favorite videos and be a part of the vote to pick the winner of the top AI generated video for this month. But let's go ahead and take a peek right now.

That was some really, really impressive stuff. I love those videos. I love the pixel art, I love the theme of the fireworks and so many of them. There was one video in there where somebody was looking out a window with fireworks and there was little, like transparent displays around there, like, feature. Oh, man, I think we've got a freaking task ahead of us to try to figure out which video is going to be the winner among those. So thank you so much for participating in that contest, sharing them in the Discord. We're going to narrow it down a little bit somehow, pick our favorites, and then let the community pick the winner. So thanks again. And if you're not in the Discord, get it. Because we're doing contests like this constantly and giving away cool gadgets, access to software and cash prizes. And by cash I mean I'll PayPal you. We're doing some really cool stuff in the Discord. You guys do impressive work. We've got a talented community in there. So thank you again for submitting. That was awesome. I am going to be at south by Southwest out in Austin on March 12th and 13th for this content Hacker Live event. We're doing a whole VIP experience on March 12th where myself and a handful of other really smart futurist people are all going to be on stage for hours sharing our thoughts on content creation and the future of marketing and the future of AI and where all of this is heading and how you can use it in your business and your content creation process. It's going to be a really, really fun night. And if you're out in Austin for South by Southwest, you're going to want to attend this. If you're thinking about going to Austin for South by Southwest, make sure you get yourself set up pretty soon because those hotels do book up quick again. The event's called Content Hacker Live. I'll make sure the links in the description if you're interested in learning more about it. It's going to be a fun evening and I'm hoping to meet some of you in person, but that's what I got for you today. Thank you so much for tuning in. If you haven't already, check out Future Tools. I curate all of the coolest AI tools that I come across on this website every single day. I curate all of the most interesting news in the AI world and I have a free newsletter where I will just send you the coolest tools and the most important news in the AI world every single week. And when you sign up, I'm going to give you free access to the AI Income Database. It's a whole database of interesting ways to use AI to generate some side hustle income. Just a cool little free bonus for joining the newsletter. You'll get it immediately after signing up and you can find it all over at FutureTools IO. Thanks again for tuning in. If you love staying in the loop with AI news, AI tools, you want some AI tutorials, make sure you like this video and you subscribe to this channel because I'm making videos like this constantly. I'm doing my best to keep you in the loop, keep you informed, keep you educated, and I just love nerding out about this stuff. So once again, thanks for tuning in. I shoot these on Thursdays, release them on Fridays, so there's probably AI news that I missed, but if I did, I'll try to get it into next week's news video. I really, really appreciate you. Thanks again to HubSpot for sponsoring this video and I'll see you all in the next one. Bye Bye.