AI News: Code Interpreter, SDXL, Claude 2, X AI, Pika Labs, and More!

So of course, the busiest week that we've probably seen in the AI world since March of 23 just so happens to be the week I'm on vacation in Colorado. So that's awesome. But don't worry, I've managed to keep up with all of the things that are happening, and I'm going to break it all down for you right now. And if you're not familiar with what AI is, well, Vice President Kamala Harris has a great explanation for you. AI is kind of a fancy thing. It's. First of all, it's two letters. It means artificial intelligence. Now that we've gotten that out of the way, let's, let's talk about Code interpreter. Last week, OpenAI announced that code Interpreter would be available for all Chat GPT plus members. And this week we finally all got access to it. The cool stuff that people have been doing with it has been all over Twitter lately. Now, I do plan on making a full breakdown of some of the coolest things I've seen Code Interpreter do, but for now, here's some examples of what some people have used it for. My buddy Amar here actually made a video game where Elon Musk fights Mark Zuckerberg. You using Code Interpreter. We could check out the results here where Elon and Zuck fight. And this was all done in what he says took only 20 minutes to code up. He said he asked GPT4 code interpreter to make an Elon Musk versus Zuck cage fight game and it pulled it off in 20 minutes. It goes on to say that Midjourney generated the sprite sheet to actually get the images inside of the game. Here's an example from Skull Skip showing off that Chat GPT with Code Interpreter can actually detect a feature faces. And here's another example from Chase Lean where he has this really wide image. He uploaded it into Chat GPT with Code Interpreter and managed to get Code Interpreter to actually pan across the image for him and make this really cool animation. Now, to get access to Code Interpreter, first off, you have to make sure you are a Chat GPT plus subscriber. Come down to your little three dots settings here in the bottom. Click on Settings, click on Beta Features and make sure that Code Interpreter is enabled here. Once you've enabled it, underneath the little GPT4 button here, you should see the ability to use Code Interpreter. Now another thing that Code Interpreter is really good at is interpreting data that you upload from something like a CSV file. So here's an example that I created where I made a column with dates, a column with foods that I ate, and then a column with how I felt after eating that food. Now this is all just fake example data for demonstration purposes, but if I export this here as a CSV file, I could come into chat GPT with code interpreter enabled. I could drag and drop this CSV file into code interpreter. I could then enter a prompt like this Here's a CSV file of foods that I ate and how I felt after eating them. Please analyze the data and find correlations between the types of food I ate and how I felt afterwards. You could see it starts by duplicating the rows inside of ChatGPT here and to show you an example of what it's finding in the CSV file, it then generates another spreadsheet here with analysis of the various foods and how I felt after eating the various foods, then found the correlation. It says, interestingly, all foods that are labeled as fried have been associated with a headache. So it correctly identified that any foods labeled as fried foods are the foods that resulted in the headache. So it properly found the correlation that I was hoping it would find. Also this week, Sarah Silverman, along with a handful of other authors, are suing OpenAI and Meta for copyright infringement. They're claiming that the books that they published were actually scraped by these companies for use in their large language models from sites that acquired them illegally, like torrent sites. And the claims, the authors say they did not consent to the use of their copyrighted books as training material for the company's AI models. Also this week we get a sneak peek at stable diffusion XL 1.0. Previously we had the ability to use SD XL 0.9 over on sites like Clip Drop, but right now we can actually use Excel 1.0 inside of the Stable Diffusion Discord. In fact, if we jump into the Stable Foundation Discord here, the way it works is we can prompt an image, it will give us two images and then it asks us to vote which one we like better, A or B. This is helping to train the new Data set for SDXL 1.0. Inside of the Stable Diffusion Discord. We would create a prompt by typing slash dream and then our prompt. So in this example a gray wolf in the woods surrounded by snow. You can see it then gives us two options to choose from, option A and option B and then we tell it which one we actually prefer. I prefer option B. I can click vote B and I have just helped train the algorithm to improve stable diffusion XL 1.0. We also have the options to redo, resize and restyle, in which case it added the style isometric and now we have two new options to choose a favorite between and which I think I like option A better than option B. Let's go ahead and vote A and once again I just helped improve the SDXL 1.0 data set. I will make sure it's linked up below so you can jump into their discord and play with it yourself and start seeing what SDXL 1.0 is capable of. In some of the biggest news that came out this week, Anthropic released Claude 2 into open beta. So anybody in the US and the UK right now can use Claude 2. It's also available via API for specific companies. I am going to do a deeper dive video on Claude 2 so I'm not going to go into too much depth on this video, but some the highlights of what Claude 2 is capable of is that you can input up to 100,000 tokens. That's roughly 75,000 words between input and output of what you can get from this Claude 2 model. That makes it the largest publicly available model that we can use right now. And I've been using it quite a bit, quite honestly for a lot of things. I prefer it over chat GPT and GPT4 at the moment, and you can use it completely for free. While it's an open beta right now, simply go to Claude AI chats, it will ask you to log in and you can start using it for free if you're in one of the countries that it's available in right now. Similar to Code Interpreter, we can also upload files and have it work with those files. So for example, if I upload this same food test file and give it the prompt, find correlations between the foods that I eat and how I feel after eating them, you can see that similar to Code Interpreter, it finds the proper correlation based on the data you provided, there seems to be a correlation between eating fried foods and getting a headache. I can also upload entire PDF files like this long net scaling transformers to 1 billion tokens paper and have it actually break down what this paper is about. I can give it a prompt along with the PDF, explain this research paper like I'm five. It gives an explanation of what this research paper is all about and then at the end it says in summary, the research has created a new machine learning model called Longnet that can understand extremely long pieces of text. Longnet does this using a technique called Dilated attention, which focuses on parts of the text at a time and by using a special distributed training method, Longnet performed better than other models at understanding both long and short texts. So as of right now, Claude 2 does a lot of the same stuff that we can do with chat GPT, with GPT4 and with code interpreter turned on, but it's completely free for anybody to use if you're in the US or uk. And quite honestly I've been very impressed with the results that I've been getting out of Claude 2. Again, I am going to do a longer breakdown on Claude in a future video. Make sure you're subscribed to the channel to make sure you can catch those Also this week the email autoresponder company Beehive, the company that I actually send my weekly Future Tools newsletters through, announced that they were adding a bunch of AI features into their platform. They've added an AI writing assistant, AI text tools to shorten, extend or change the tone of voice, an AI image tool to actually generate images straight with inside of your newsletters, an AI translator to translate your emails into various languages like Spanish, French, German, Italian, Portuguese and Brazilian Portuguese and a bunch of other really cool features. In fact, here's the newsletter that I'm working on that by the time you're watching this video has likely already been sent to you. But you can see that I can select a section of text here and there's this little button here that says AI Writing Tools. If I click on this I can improve writing by changing the tone of voice to any of these. Let's switch it to a friendly tone and you can see it just rewrote it for me in a different tone. Hey there, guess what? Stable Diffusion has just released the brand new SDXL 1.0 model and it's now available for all the awesome folks in their Discord community. I like the way I wrote it better, but you get the idea of what it can do. If I scroll down a little bit here, you can actually see this is an image that I generated using Beehive's AI image generator. And I mean it spelled you wrong, but I'm actually surprised that it got the text fairly decent. So if you have an email newsletter, this is not sponsored by Beehive whatsoever, it's just the tool I use. They added some cool AI features, so maybe it's one you could check out if you have or want to send a newsletter. Also this week Shutterstock expanded their partnership with OpenAI and they signed a six year agreement to provide high quality training data. So essentially what this means is that OpenAI will be able to train their image models like Dall E on images sourced from Shutterstock. And in exchange, Shutterstock gets to use Dall E models with inside of their platform. It's sort of a win win for both. Shutterstock is gets AI image generation. OpenAI gets training data to train their models on. Now, Shutterstock could be trying to fight this AI revolution because it is a serious threat to their business model. If anybody can generate AI images, why would they go and source them from stock photo websites? But instead Shutterstock is choosing to embrace and work with AI and actually evolve their company, which I think is a smart move for any company that's selling stock photos. Because quite honestly, it's going to be a much harder and harder sell for stock photo companies to actually make money off of selling stock photos. If anybody could just generate anything they can imagine. Also this week Google announced Notebook lm. This was originally called Project Tailwind and essentially what it is is you could put your files inside of your Google Drive and it will turn all those files into a searchable chat so you can essentially have conversations with all of the documents and information inside of your Google Drive. They're starting to roll out beta access to users now, and this is something I'm personally very excited about, but haven't had an opportunity to test. But I guarantee we'll be making some more videos about this in the future as well. Now here's a really fun tool that was released this week as well called Pika Labs. And what this is is it's actually a text to video generator similar to Gen 2 or model scope or zero scope. However, it works really well with image to video. You can see the example input that they have here of a still frame of a woman here and then they put a prompt of a girl in the wind and you can see she's blinking now and her hair is blowing. Here's another one of a cowboy still image. And now you can see the wind blowing and the shirt moving around. Now it's not publicly available to everybody yet, but you can join their beta by filling out a quick type form survey here. And they're seemingly pretty quick to give people access. Once you have access, it's available inside of a Discord bot and to use it you simply type create. If you click over on this plus one more here, you get this option to upload an image so you can get this image box. Let's upload this picture that I made of myself holding a cup of coffee and then let's add the prompt. A tired man staring at a steaming cup of coffee. Now, one of my complaints with this Pikachet is they only have two generation rooms and it just moves really fast, so stuff gets lost really quickly. So one of my pieces of feedback to that would be maybe add some more rooms because it's getting a little hectic in here. I just downloaded the video and I'm watching it on loop and you can see that I've got a subtle head movement up here, and you can see some steam rising out of the coffee that I'm holding. And it almost looks like maybe it's raining in the background on the windows. So really cool. This only took about 30 seconds to generate. I started with an image, I added the prompt you just saw, and now I've got an animated video based on that initial image that I started with. And just like Mid Journey, you can actually scroll through the feed here and get some ideas and inspiration from what other people have generated. Like, here's a cool video where somebody animated a little Pegasus image, and here's one where somebody animated the dark side of the moon log. So really cool, fun technology. You can find it over at Pika Art. I'm excited to see how this one evolves, and I'm gonna be playing with this one a lot more and putting it through its motions because I really like the idea of starting with an image that I generated in like Mid Journey or Stable Diffusion, and then pulling it into a tool like this to animate it. Now, as a video creator, this is an announcement that I found really cool as well. It's from Epidemic Sound and it's called Sound Match. Essentially what it is, is you can upload a video, select that chunk of video, and it will analyze that video and then find the perfect background music from the Epidemic Sound library to go along with that video clip that you just uploaded. Now, because so many people have put videos on YouTube and added background music from Epidemic Sound and then gave attribution to Epidemic sound on those YouTube videos, epidemic sound knows what type of music goes with what style of video, and it's using that as the training data to figure out what music is going to work well with your videos. I personally haven't tested this out yet, but I'm excited to play with it. I don't normally add background music to my videos, but with a tool like this that makes it that easy I might start. Also this week, Bill Gates wrote up an article about the risks of AI. I'm not going to go into too much detail about it. I do recommend reading it. Essentially his thoughts on it are yes, there are a lot of risks to AI for scams and deep fakes and potential for harm, but also with AI we can use the technology to also prevent those types of things. And he also talks about how yes, a lot of jobs will be obsolete because of AI, but also a lot of new jobs will be creative and that humans are innovative and will find new ways to add value to the world and earn money in the world. The overall message is yes, AI is scary, but we could also use AI to mitigate the scariness of AI. I'll link it up below so you can read it. Definitely worth the read. At the very least, copy and paste it into Claude 2 and have it summarize it for you. Another big announcement that happened this week, but also a fairly cryptic announcement, is that Elon Musk announced X AI. He's working with some of the top AI researchers in the world. People from DeepMind, OpenAI, Google Research, Microsoft and Tesla and the University of Toronto are all teaming up to help develop this new xaion. Now, we don't have a lot of information about what XAI will do other than the goal of XAI is to understand the true nature of the universe. We'll learn a lot more about Xai on July 14 from a Twitter spaces. Unfortunately, when this Twitter spaces is happening I'll be 30,000ft in the air flying home from Colorado. But I will try to get the notes or listen to the recording or replay after the fact so that next week I can update you a little bit more on what the heck is xai. Now, this isn't a sponsored video, but one of my regular sponsors of this channel, Shopify, actually announced a new AI feature for their platform as well. We can see in their example here that there's a new sidebar window called Sidekick. You can ask it questions and it will use your sales data to answer those questions. For example, I've had a drop off in sales for March to now. Why it gives an answer. The dip in sales is probably due to minimal snow. These are summer months in many regions, so it's natural for sales to to decrease. So it actually looked for correlations between their sales and the time of the year. They then give it the prompt put everything on sale and it says all right, let's get that sale started and it gives a 10 sale on all the products to increase sales during the off season. They then give the prompt add my new surf collection to the homepage and it says, all right, let's add that collection to your theme. They click on preview and now you can see a surfboard theme on their homepage. Remove the snowboards and make my store look like a surf shop. Write new content for my Hero banner focused on selling surfboards. It basically generated a new page, turned it into a surf shop instead of a snowboard shop. You'll definitely be hearing a lot more from me about this as I try to build out my own Shopify store. Again, Shopify didn't sponsor this video, but these tools are going to be super handy in the future as I build out the future store. Also this week Google's Bard announced a whole bunch of new updates and this may be in response to Claude2 and code interpreter since those two big announcements came out this week and Google is maybe trying to keep up a little bit and they put out this pretty blog post explaining some of the new features. But I find this news update here over@bard.google.com updates to be even more helpful because it's just a bulleted list of what they added. Bard is available in new places and languages. They've got over 40 new languages including Arabic, Chinese, German, Hindi, Spanish and more. They added Google Lens to Bard so you can now upload images alongside text in your conversation with Bard. Bard can now read and respond out loud. So it's got text to speech to actually read to you. It's now got pinned and recent threads so it's it can actually remember some of the conversations you've had in the past. It says you can now pick up where you left off on past Bard conversations. You can now share your Bard conversations with others. This was something that we were able to do in Chat GPT for about a month now. Now Bard has this feature as well where we get a special link and can send people straight over to our conversations. There's new responses so you can make the responses simpler, longer, shorter, more professional or more casual. And you can now export your Python code to Replit. So all of these new features just rolled out in Bard. Again, this is something that I need to go deeper on in a future video because all of these LLM chatbots are getting so much better between Claude 2 chat GPT with code interpreter and now all of Bard's new updates. I think it's time that we do a nice comparison video where we put them all to the test and pit them against each other. So something I'll definitely be doing in the near future. Let's go ahead and click this plus Button to upload a file. Looks like we can upload JPEGs, PNGs, and WebP files. Just for fun, let's go ahead and upload this image from my last news YouTube video. And let's give it the prompt. Describe what is happening in this image. And it says, sorry, I can't help with images of people yet. So I guess that's not the best example. Let's go ahead and upload a different file. I'll upload this little image of a town that I made for, like an anime demo I made in the past. Describe what is in this image. So just so you get a better idea of what the image actually is, this is what we can see here. It's like a colorful anime town. I think there's like a person in the background, there's some street posts, etc. The image you sent me is a cartoon illustration of a city street. It shows a woman riding a bicycle, maybe on a quiet street at night. Definitely daytime. The street is lined with tall buildings and there are a few cars parked along the curb. Definitely no cars on the curb. The woman is wearing a long coat and hat and she is carrying a backpack. So it's basically hallucinating what it sees in this image. Let's just be real. Let's try the prompt. What colors do you see in the image? The main colors in the image are black, white, dark blue, and red. Okay, maybe, maybe so you can upload images. I haven't been too impressed with it yet, but I also haven't totally put it through its motions. I haven't given it the full rundown yet. I could be prompting this totally incorrectly again. I've been on vacation in Colorado for the last week. They just rolled this out. So I haven't had a whole lot of opportunity to deep dive and test it out to its fullest. But trust me, I'm going to. And right now the image stuff doesn't seem so great, but it might be better than what I'm giving it credit for. Also this week, Clip Drop launches Stable Doodle. Stable Doodle. You can use this for free over on Clip Drop right now. Essentially what it is is you can doodle some images and give it a little bit of a prompt like Cityscape with Full Moon and it will generate based on the scribble. And then you can also add some styles to it like photographic analog, fantasy, comic, etc. And get different styles. So let's jump on over to Clipdrop Co Stable Doodle and draw a penguin. Now I'M trying to do this with a mouse so it's not going to be great. But let's draw a body here. Let's draw head. Let's draw some wings. Oh gosh, this is really bad. I'm sorry guys. Let's draw some feet like this. And then here's a beak and some eyes and like a little puff of hair. A penguin. And click. Generate. Actually not bad. That did better than I thought it would. Let's try a different style. Let's go to Origami. This one actually looks pretty cool. So clip, Drop. Stable Doodle actually does better than I was expecting. This was my first use. Honestly, I have not tried it until recording in this moment and it only gave me two images because it says I've reached the current usage cap. But they're actually pretty dang good. It appears that if you log in, you get even more free generations. And here's what my penguin looks like with a low poly version of it. Pretty impressive. I like it especially for getting to generate some free images. One of my favorite AI art tools. This week Leonardo AI announced prompt magic version 3 as well as the ability to use stable diffusion XL 0.9. Now also if you're a paid member, you're also getting 2,000 extra tokens to test out these new tools. So let's jump over to Leonardo AI. If I come up to my fine tune model up here and click on the drop down, you can see I now have SD XL 0.9 available and for some reason I've got a wolf game character under the AI generation tool. Let's just go ahead and generate and see how SDXL 0.9 does it. That looks pretty damn good. That, that's. That's really cool. I like those images a lot. Now also they added prompt magic version 3. I can't select it because I need to have Leonardo Alchemy turned on. So let's go ahead and turn that on. And for our fine tuned model up here, let's go ahead and use Leonardo Diffusion. I like that model a lot. And let's go ahead and try version 3 beta. And let's do a wolf game character using Leonardo Fusion version 3 beta and Leonardo Alchemy which is Leonardo's own custom pipeline for generating these images. So let's go ahead and generate with that. And they look decent. This one's pretty dang good. This one's got some issues going on with it. Let's try RPG 4.0 because I really like that fine tuned model as well. And let's run this one more time. And this one actually looks pretty dang good. It's like a half human, half man hybrid, but definitely looks like a game character. Definitely something I need to spend some more time with. Haven't had a lot of chance to play with this. This announcement literally came out right before I hit record on this video. So I have done no testing with this whatsoever yet. So I can definitely generate some better images with this if I gave it a little bit more time. But I just wanted to point out that that's available so you can go and test with this and get some really cool new images from it as well. Now, last thing I want to talk about, I'm just going to touch on this really, really quickly is that last week people in the comments were mentioning that I forgot long net the 1 billion token new transformer model. And I didn't forget it. I did come across this research paper, but typically I will do separate videos on the new research that's coming out that we don't necessarily have available yet. And those are typically different videos from. Here's the latest news and announcements. This to me falls under that research category of something that's really cool that we're going to have access to eventually but not available for anybody to use yet. Now it is exciting. What this is basically saying is that we're going to get a 1 billion token limit on future large language models, meaning that you no longer have to worry about how much input text or how much output text you get back. You can write entire book series, like 10 different books, all from a single prompt with 1 billion tokens. I mean, it's just insane the types of stuff we'll be able to do with 1 billion tokens. But it's still just research. It's from Microsoft. They put out a paper on it, but there's nothing that really demonstrates the research yet. There's no way to actually use this or test it out for ourselves yet. And there hasn't been any big announcements from Microsoft or anything like that about when this might be available for users. So I haven't included it in any of the news videos, but it would have been included in one of the future research videos and likely still will be. But I don't want to go into too much depth in this video. Instead I'm going to point you to this video from David shapiro called Microsoft Longnet 1,000,000,000 tokens LLM Open AI Super Alignment Singularity Approaches. This is an awesome breakdown from David here. I watched this entire video. It's essentially a 20 minute breakdown of why we should be excited about this new upcoming research from Microsoft. So I highly recommend watching that video. I will link it up below in the comments so you can check that video out and get more information about what Longnet will eventually be able to do and why this technology is so important and powerful to the advancements of AI models. All right, that's all I got for you this week. I'm out of breath. This is, I know, one of the longer videos. Again, I haven't seen a week like this since March where it was just news announcement after news announcement after news announcement. And of course it happens when I'm still on vacation in Colorado. As you're watching this, I'm probably on a plane right now. I record these videos on Thursday, release them on Fridays. I'm flying home on Friday. So next week my next videos will be back in my normal studio. But even on vacation, I still find time to step away for a few minutes here and there and keep myself in the loop and updated with all the latest AI advancements. Hopefully I caught them all. I'm sure there's going to be some people in the comments that point out some other stuff that I missed. That's okay. I will catch you all up next week when I'm back in my studio with with all the latest AI announcements I missed. And if you haven't Already, check out FutureTools IE. This is where I update you every single day with all the coolest AI tools that I come across. This AI news page is updated every single day with all of the latest announcements in the AI world. And if you just want to keep in the loop once a week with the tldr, join the free newsletter and every single week I'll send you just a handful of news articles, a handful of cool tools that I came across, a few YouTube videos in the AI world, and one cool way to make money with AI. I send it every single Friday. It's your TLDR, your grand overview of the week in AI. If there's only one thing that you pay attention to to just kind of keep your finger on the pulse, make sure you're on that newsletter. You can find it over at FutureTools IO. All right, thank you so much for tuning in. Checking out from Colorado. Excited to be back home in my own studio next week for future videos. I've got so many video ideas that I want to make now that we've got access to Claude 2, we've got access to Code interpreter, we've got access to new features in Bard SDXL 1.0 New models in Leonardo. I mean the list goes on, on and on of videos I want to make and cool tools that I want to test. So excited to get home and make all those videos for you. And once again, I really, really appreciate you. Thanks so much for tuning in. Been a crazy news week. AI is picking up speed again and I'm here for it. Hopefully you are too. Thanks for nerding out. See you guys in the next video. Bye. By.

Sam.