AI News: The Best Chat Tool Got SO Much Better!

There has been a ton of AI news this week and I don't want to waste your time, so let's just get right into it. Starting with the fact that Anthropics Claude rolled out even more cool features, you can now create what they call projects, and a project is kind of similar to a custom GPT over on Chat GPT. You can upload a whole bunch of files for context and then every time you log into that project and have a chat with it, it's going to use that same uploaded information and every single time. You can even give it custom instructions so it knows how to handle any prompts that you enter into this specific project. They added the ability to share projects with other team members, and it also got a slight user interface update. We can see our standard chat box here, select our model here, use any of the projects we created by clicking on this little projects dropdown here, and we can see all of our recent chats here. You'll notice your profile button is no longer up here in the top right of the screen now. Instead, if you if you move your mouse to the left of the screen, the sidebar automatically pops in. We can see all of our recent chats here. You have the ability to star chats that you come back to often. We can access our projects through here as well. And if you need to change any of your settings on your account, your profile link is now in the bottom left. I can click here and access settings, change the appearance, go to the feature preview and turn the new artifacts feature on or off. We talked about that last week. It's a new sort of right side window of your screen that shows the sort of code output and things like that that you do with claude. And if you want to use Claude collaboratively with team members, you can also upgrade to a team plan over here. But let's go ahead and play with projects real quick and see what it's capable of. We can create a project up here in the top right. Name the project. I'll name it research Papers and I'll give it the description. Analyze and respond to questions about uploaded research papers and and create our project. Now we have our project window here and we can set some custom instructions. I'll tell it to answer any questions I have as concisely as possible about the research papers that are within the project knowledge. Start every response with no problem. Matt, we'll go ahead and save the instructions here. I have a handful of research papers that I've downloaded and saved. Let's go ahead and grab A few of them dump them into our knowledge base like this. You can see I've uploaded six PDFs here of various research papers and and I've used 80% of the knowledge size. They give you a maximum of 30 megabytes. It's not based on how many files, but the size of the overall knowledge base you're uploading. Let's go ahead and ask what papers are available inside my project knowledge? It starts with no problem, Matt. The papers are available inside your project knowledge are generative agents, interactive simulacra of human behavior, game based platforms, et cetera, et cetera. And we can say what correlations can you find between these papers? No problem, Matt. The main correlations between These papers are AI and machine learning. All papers are focused on advanced AI and machine learning techniques. Generative models. Papers 1, 4 and 5 deal with generative AI models, particularly diffusion models. Papers 1 and 2 emphasize interactive AI models and it found the relationships between these five documents that I uploaded. Now what's cool is if I come back to my homepage on Claude and I want to use that same set of knowledge that I've already updated, I I can just click Use Project click on the research papers and now you can see that it is set to the Research Papers project and any prompt that I give it here will reference that knowledge that I already uploaded as well as the custom instructions that I gave it. This to me seems really handy, especially combined with the Artifacts feature which I showed off in last Friday's video. But just as a quick reminder of this new Artifacts feature, I can type something like Create the HTML for a sales page about surfboards. And you'll notice it opens up a a new right hand side of my screen where it actually writes all of the code. I can continue to have conversations on the left while the code is generated on the right and I can actually preview the website it generated right here on the right. Flip between the code and the website preview without ever actually leaving. Claude, I don't know about you, but I really feel like these new features in Claude make it really, really useful. Google also made a bunch of updates and announcements this week, including the fact that Gemini 1.5 Pro Pro has a 2 million context window that's a combined 1.5 million words between the amount of text inputted and outputted back to you. The entire set of Harry Potter books is about 1.1 million words, and the entire set of Game of Thrones books, A Song of Fire and ice are roughly 1.7 million words. So you can almost upload that entire series of books and and have conversations with it. That's how large this model is. It's basically to a point where context windows are no longer really an issue. If you're willing to use Gemini 1.5 Pro right now. Now Gemini already let us have a 1 million token context window and I would say probably 99.9% of people didn't even need that large of a context window. A 2 million context window, while a huge leap and insane that we can do it, probably still not going to be applicable to most people, but still a really killer feat. Gemini 1.5 Pro can also execute code now as well, and the Gemma 2 open source model from Google was also made available this week. And based on the benchmark tests here, we can see that this new Gemma 2 model, both the 9 billion parameter model and the 27 billion parameter model outperform llama 3 in most benchmarks as well as Grok in most benchmarks. This week Google also announced that 110 new languages are coming to Google Translate. They're using AI to add these new languages, including Cantonese, Nko, I don't know how to pronounce that, and Tamazite. It was also announced this week that Google is rolling out their sidebar inside of Gmail. We can see their example here. The Gmail window is on the left and on the right Gemini pops out and offers the ability to summarize conversations, show me emails that are unread, draft an email, or get more suggestions. But basically it opens up a chat window that can interact directly with your Gmail emails. It looks like in order to use it you've got to be a Gemini Business or enterprise, Gemini Education, or have the Google One AI premium. And we can see that the rollout's already started. I checked my own Gmail, I don't have it yet and I am a Google One subscriber, so as soon as I have it available, I'll show it off. Moving on to OpenAI News. It was announced this week that that advanced voice assistant, you know, the one they demoed several weeks back that they got in trouble for because it sounded a little too much like Scarlett Johansson, but everybody thought the feature was really cool. Well, they've decided to push it out even further. They say they had planned to start rolling this out in alpha to a small group of ChatGPT plus users in late June, but need one more month to reach our bar to launch. They're improving the model's ability to detect and refuse certain content they're also working on improving the user experience and preparing their infrastructure to scale to millions of while still being able to maintain real time responses. However, it does appear that some people have been starting to get access to it. I came across this X post here that shows that it even generates background sound effects while narrating. So it appears that somebody asked it to tell it a story and in the story actually included sound effects. One night a storm rolls in. Thunder, lightning, the whole shebang. A couple of teens trying to be brave or just plain stupid decide to check out the house during the storm. They sneak in through a broken window and start exploring with nothing but a flashlight and their phones for light. Pretty crazy as it's telling the story you hear like thunder going on in the background. OpenAI also announced that they trained a new model based on GPT4 called Critic GPT. It's basically designed to catch errors in ChatGPT's code output. We can see in their example here that they're asking it to write a Python function. It writes the Python function and then Critic GPT then reviews the Python function and points out any issues with it. You can see that in this specific function it points out that checking if an absolute path of a file is within a directory by using Startswith is insecure. Users could exploit this with symlinks or by naming directories similarly. Now it does say that Critic GPT suggestions are not always correct, but we find that they can help trainers to catch many more problems with model written answers than they would without AI help. Now, I'm not sure exactly when this is going to be available for everybody to use. In their blog post here they say we're planning to scale this work further and put it into practice. The Mac Desktop app is now available for all users. I did actually make a tutorial on how to use this desktop app if you want to learn how to use it yourself. If you were a PLUS user, you've probably already had access to it and been able to use it. But now if you're a free ChatGPT user, you can. You can now download the Mac app and use it as well. If you press option + space bar, it opens up a little chat box, kind of like the Spotlight window that you already have in your Mac. Except whatever you type into it will ask the question or give the prompt to ChatGPT. And speaking of ChatGPT and their desktop app, OpenAI actually purchased another company this week. They bought a remote collaboration platform called Multi. If you're not familiar with Multi, it's a collaboration software where somebody can actually share their screen and work collaboratively on the same desktop. You can see it's got shared cursors and drawing, simultaneous screen sharing. You can click and type into somebody else's app. And now this is owned by OpenAI, so it's not too hard to see where OpenAI is likely going with this. We might see a future where when we run into a problem on our computer, we hand over control to ChatGPT and it views your screen and then collaboratively solves the problem with you. Imagine this if you're a coder as well and you're having ChatGPT write the code and then you're opening up the output of the code and it's not working properly. Well, ChatGPT will likely in the future be able to see your screen, offer suggestions and point out solutions, and likely even iteratively update the code based on what it's seeing on the screen. Now, I'm just speculating here, but to me this seems like the logical conclusion of where OpenAI is probably going with this. Toys R Us this week, which I didn't even know still existed, leveraged OpenAI's Sora to make a brand new Toys R Us commercial. But here's a peek at what the actual ad looks like. Did you ever wonder how Toys R Us and Jeffrey the Giraffe came to be? The son of a bike shop owner, Charles Lazarus had a vision.

That would go on to change toy stores forever.

Now, that's just a snippet of the commercial, it's not the whole thing, but opinions have been very mixed on it. A lot of people hate the fact that companies are using AI to create commercials, and a lot of people are talking about how maybe this is the future of commercials. Also, it hasn't been claimed that a hundred percent of this commercial was made with AI, but we do know that a large majority of it was created with sora. And then I'm sure somebody came back in and did some editing and cleanup on it. Today's video is sponsored by LTX Studio, which is perfect because they recently recently rolled out a bunch of new updates. This new pack of updates is called Visions. This update gives you a ton more flexibility over creating your videos, like the ability to completely start from scratch with a new video or paste in a script and let LTX Studio make the video based on your script. With my script broken down into scenes, I can simply click start and in less than a minute I have a complete storyboard created from the script that I just uploaded, but check this out. Maybe I want to turn my whole video into a completely different style. As part of the New Visions update, I have the ability to drag and drop an image like this into the style reference in LTX Studio and then watch as it reworks my entire storyboard to match that new style that I just uploaded. Once I'm happy with every single shot in every single scene of my video, I can generate each video clip individually or I can come up here to export and click Generate Missing Shots to generate the entire video all at once. They also rolled out a preview pitch deck feature with this latest update where you can export your entire storyboard and it will turn it into a beautiful pitch deck that you can use to sell the story that you're creating here. And this is just scratching the surface of what you can do with LTX Studio. It'll also create voiceovers, create a musical soundtrack, allow you to use a consistent character throughout your video, edit the location, the lighting and the weather of every single scene and and so much more. It's actually a pretty impressive suite of AI video editing tools. You can join the waitlist for LTX Studio today by clicking the link in the description below. They're accepting users now and thank you so much to LTX Studio for sponsoring this video and even more OpenAI news they partnered with Time Magazine on a strategic Content partnership time and OpenAI announced a multi year content deal and strategic partnership to bring TIME's trusted journalism to OpenAI's products, including ChatGPT. OpenAI will gain access to current and historic content from TIME's extensive archives from the last 101 years. The partnership will also enable Time to gain access to OpenAI's technology to develop new products for its audiences. It is very interesting to me how the sort of media narrative has really shifted lately. Obviously the New York Times are still in the middle of a lawsuit with OpenAI, but almost all of the other media companies are coming out and saying we're partnering with OpenAI, we're partnering with OpenAI. But in the early days most of these media companies were really, really resistant to this AI technology, saying why would people come and read our websites anymore if they can just go talk to ChatGPT? We know that ChatGPT most likely trained on all of this data from all of these sites and now ChatGPT is getting ahead of even more potential lawsuits by partnering up with a lot of these companies that create this content so that there is still some sort of incentive for all of these companies to continue to create content, report on the news, do deep dive research, and even if people find that information through OpenAI, it's still in some ways going to benefit the media House because OpenAI is apparently licensing this content in some way. And that wasn't the only partnership that Time made. This week Time also partnered with eleven Labs, the company that does really realistic text to speech. Apparently Time is trying to go audio native. Basically, it looks like if you go to the Time website, they're going to have embedded AUD players with realistic voiceovers that you can just click play on and it's going to sound like a real human is reading that Time article back to you. Here's an actual example that's live right now on the Time website. If I scroll down we can see this player here and if I press play headline We Are the World Power How Joe Biden Leads by author Massimo Calabresi, Washington. You can read the transcript of the interview here and the fact check here and we can see in the corner it's got the 11 labs linked and in the future I think most websites are probably going to go this direction, not just Time. I think even like WordPress, blogs and things like that, you're going to see a lot more and more of them integrate tech from companies like eleven Labs. So you have the option to either read the article or blog post or listen to it in a realistic voice. And eleven Labs even allows you to do things like train your own voice into the system. So when I create blog posts, you might be able to go to that blog post, press, play on it and hear the blog post as if it was read by me, which is pretty cool. That's not the only news we got out of 11 labs. This week. They also introduced a new reader app where you can listen to any text on the go with the highest quality voices. This 11 Labs reader app lets you listen to any text content on the go. So now any article, PDF, ePub, newsletter or any text on your phone can now be turned into an audio that you can listen to with a realistic voice. The app is available right now on iOS and coming to Android really soon. I haven't played around with it too much yet, but I am excited to use this to listen to a lot of the news articles and research papers that I've been meaning to get around to reading. Also this week Character AI rolled out a new feature that allows you to talk to the AI avatars. Now Character AI is a very interesting company. They are huge, especially among the younger demographics. Character AI is probably the most popular AI tool for, like, younger generations. But they just rolled out this feature where you can have phone calls, essentially with fictional AI characters. If, for example, I wanted to have a conversation with Spider Man, I can try with Voice. Ask it something like, what's your favorite food? Oh, man, that's tough. I love a good pizza. That's pretty much the standard. But I'll also never say no to dumplings or just generally spicy stuff. How about you? Now, this voice feature isn't the new feature. The new feature is that you can actually call and have this conversation over the phone. You can see here, when I hover over this little phone icon, it says calls available on the app. I'm not going to download the app to demonstrate it because I'll probably never use it again, but you get the idea. Same concept. Just like you're on a phone call instead of in a text chat. While we're on the topic of AI and audio, all those fun AI music tools that we've been playing with lately, you know, Sony and Udio and things like that, well, the record labels are starting to sue those companies, claiming that they were trained on copyrighted music that they didn't have the right to use. Now, this is sort of like a muddy area because they're going to claim that that was fair use. The record companies are going to fight that it was a massive infringement on copyright, but there really is no precedent for this yet. So it's going to be interesting to see how it plays out, if you want my opinion on it. I feel like the record labels are going to really, really be fighting for relevance in a world where anybody can kind of produce their own music these days, and record labels actually do less and less of the marketing for musicians. I feel like record labels are losing relevance really, really quickly, and some of these lawsuits are kind of them trying to stay relevant and grasp for the power and the control that they currently have. But it's going to be something fascinating to watch and see how it plays out, because it is another area where I can kind of see both sides and understand the arguments on either side. YouTube, on the other hand, is trying to avoid this exact scenario. This week it was announced that YouTube is trying to make AI music deals with the major record labels and get in front of this. According to The Financial Times, YouTube is offering to pay Universal Music Group, Sony Music Entertainment and Warner Records lump sums of cash in exchange for licensing their songs to legally train its AI music tools. This apparently isn't for their existing dream track. Feature, but it's going to be used to train new AI tools that YouTube is planning to launch later this year. Again, this is a flip side of the coin from the last article, which is going to be interesting to watch to see how this one plays out. Because if these record labels are willing to sign some of these licensing deals to allow music to be trained, well then that could be the solution for companies like Suno and yu. Do go and agree to terms and get the licensing and make all this go away. We'll see how it plays out. It's going to be interesting to watch nonetheless, and I'll do my best to stay looped in and tell you what's going on as this plays out. This week, Figma had a big event called the Config Conference and they made a whole bunch of AI announcements around what you can now do inside of Figma. It also got a complete redesign and overhaul. Figma's now going to be able to take text prompts and create designs from those text prompts. They showed an example of how Figma could create an app design for a new restaurant. A few seconds after they typed the prompt, Figma mocked up an app with menu listings, a tab bar, and even buttons for delivery partners like uber Eats and DoorDash. They claim it was a pretty generic mobile mockup, but it was better than starting from a blank slate and you were now able to edit and reconfigure the design based on the template that it created for you to start with. It can also add auto generated text instead of using the generic Lorem Ipsum placeholder text that people tend to use on websites. And it has a new AI enhanced asset search to better help you find the various assets for designing your site. Now right now they're rolling this out as a beta feature and it's free for Figma users in the beta mode, but eventually it is going to be a paid feature. The company Synthesia announced a handful of new updates this week. They're a company that makes these like AI avatars that are kind of designed for having, you know, AI sales bots and things like that. Supposedly the avatars are more expressive, showing excitement and doubt and fear and anger and things like that. They've also got what they call personal avatars, which can actually look like you and use your voice and sound like you, but also speak in other languages. They're also adding the ability to do full body avatars instead of just, you know, from the neck up and a handful of other features. I'll link the entire YouTube video up. If you want to learn more about what Synthesia is up to, a new AI video editing tool rolled out this week from the company Captions. You can use this tool and feed it a standard video and it will edit the video with different backgrounds, make it look like you've got a green screen, add text like you're seeing on the screen right now. Add stuff that looks like after effects animations. It looks like a pretty cool tool. Now, I haven't played with this one myself yet, but I think this one might deserve another video in the future. If it turns out to be a pretty cool tool, look out for that video later on and we'll see how it does. A few months back, I talked about a tool called Magnific, which is this really good upscaler tool, but it kind of hallucinates extra stuff into the image and sort of adds this creative flair to the image. Well, they just rolled out this new relight feature and it lets you take your images and sort of reimagine. Like for example, this thumbnail here of me holding up all these glasses. I reimagined it with me in outer space. Simply uploaded image right here. You have the option to transfer from a prompt another reference image or a light map. But let's just go ahead and do. From a prompt, I uploaded an image of people dancing. And let's just add people dancing underwater. I'll leave everything the same, although we do have some other style options and things. And if I click relight, you can see it. Relit it like this. Here's the original image, here's the new image that it made. It does a really good job of completely restyling the image. This week, stability AI got a new CEO who is the former CEO of Weta Digital. If you're not familiar with Weta Digital, Weta is a visual effects company that's worked on films that you might have heard of. You know, things like Avatar, Planet of the Apes, the Marvel movies, Transformers, Game of Thrones, and is probably best known for a little series of movies called the Lord of the Rings. Things keep getting worse for the rabbit R1. This company called Rabbit Tude, which is essentially a company who's trying to poke holes and sort of find flaws in the rabbit R1. Well, they uncovered a huge data breach. The team gained access to the Rabbit code base and found hard coded API keys in the code which would allow anyone to read Every response every R1 has ever given, including ones containing personal information brick all the R1s alter the responses of all the R1s and replace every R1's voice. They found the API keys for 11 labs, Azure, Yelp and Google Maps in there. On top of all of the bad press and all of the bad reviews the Rabbit has received, it also turns out they did a pretty poor job with code and left a bunch of API keys exposed, meaning that anyone that has the rabbit R1 is at risk of having their data leaked. A company called Etched announced a new chip which is apparently the world's first specialized chip for Transformers. The Transformer is pretty much what all large language models that you use are built on top of. It's also what so Gemini Stable Diffusion and all of these tools use. And these chips are more than 20 times faster than the Nvidia H1 hundreds, which is what most of the AI companies these days use to train their AI models. This looks to be pretty impressive when they become available. And they also appear to be a competitor to Grok the Groq Grok, which I just so happen to be holding a pen for. And the day they announced this, Grok announced that they've achieved 400 tokens per second with the Llama 370B model. This message appears to be a direct response to the etched comment. As pointed out here by Akshay Meta started testing user created AI chatbots on Instagram. We've seen some of these AI chatbots with like Snoop Dogg but under a different name and Mr. Beast under a different name and I think Tom Brady under a different name. Well now they're starting to roll it out so individuals can create their own custom chatbots. It appears to be rolling out an early test in the US of the AI studio so you might start seeing AIs from your favorite creators and interest based AIs in the coming weeks on Instagram. They will primarily show up in messaging for now and will be clearly labeled as AI. And finally with the Olympics coming up, they are going all in on AI this year, including an AI version of Al Michaels. People can get customized feeds of what they're interested in from the Olympics and it will be commentated on by Al Michaels. But it's AI Al Michaels. Here's a quick example of what that might look like. Ty Kelly welcome to your daily Olympic Recap. Your personal rundown of yesterday's most thrilling Olympic moments.

Since you're a swimming fan, let's head right to the pool. Team USA secured a stunning victory in the men's 4x100 meter medley relay, smashing the world world record over at the diving venue, Krista Palmer showcased resilience and skill, overcoming past knee surgeries to qualify for the women's springboard final. I'll go ahead and stop it there. You get the idea. You're going to be able to tell it what sports you're interested in. And it will create a custom little video of a recap with Al Michaels commentating on that recap, even talking to you by name if you want to. And that's what I got for you. A lot of AI news came out this week. It was weird because it kind of felt like a slower week to me, but as I was rounding up all of this AI news, I was like, holy crap, there's a lot going on. It's gotten to the point where 30 new AI announcements in a single week feels like a slow week, which is just wild. But anyway, if you enjoy videos like this and you want to stay looped in on the AI news, give this video a thumbs up and make sure you subscribe to this channel. I'll make sure that videos like this keep on showing up in your YouTube feed. And if you want to find more AI tools to play with and stay looped in on the AI news on a daily basis, check out Future Tools IO, where I keep the news up to date every day. I'm adding cool AI tools as I find them, and I've got a free newsletter. You join the free newsletter, I'll hook you up with the AI Income Database, a database of cool ways to make money using these various AI tools. And you'll be entered to win cool prizes in our monthly competitions, which is exclusive to people who subscribe to both the YouTube channel and the newsletter. So again, head on over to FutureTools IE to find all that stuff. Thank you once again to LTX Studio for sponsoring this video. And thank you so much for tuning in, nerding out with me, and keeping your finger on the pulse of the latest in the AI world. Thank you so much once again. I really appreciate you. I'll see you in the next video. Bye. Bye.