My Favorite Claude Update, OpenAI's Sora Leaked & More AI Use Cases

We haven't had many big announcements in the AI space recently, but we're getting features left and right and some of these I'm absolutely loving. These are the things many Gen AI power users have been requesting, because among this week's updates we got a few features that I've been longing for since forever, and I know that a lot of other power users have been requesting these too. So we're going to be looking at the brand new cloud custom styles feature, OpenAI Sora Leak this week we'll be reviewing that story. Plus there's further updates to the ChatGPT app and so much more. In this week's episode of AI News, you can use the show that rounds up all the various updates and releases in AI space and presents them to you in a way that focuses on using these tools rather than something that might be available one day. Let's get into it. Okay, so first off, we have Claude introducing two features here. One of them is pretty straightforward. We have this in a lot of other LLM platforms. Instead of just attaching files, if you hover over this now you can also attach files directly from Google Drive. This is just a little convenience feature, but the big one here is this, and I want to spend some time on this because I think this is actually significant and it's only been out for a few days, but I personally love using this and it has very quickly become one of the reasons why I often started defaulting to Claude. And it's for the ability to save a preset style. Okay, it comes by default with normal, concise, explanatory, formal. And I added this tech mentor style here and I'm going to show you right now how to do this for yourself in the best way possible. And adding a style like this is not something brand new. This is something you could do inside of ChatGPT. You could do it in here with the Projects feature. That's exactly what I did. I had Igor's email writer here set up and it basically came with these custom instructions where I told this to imitate my style, tone, voice and structure of the following examples. And then I gave it five examples of emails that I wrote myself back by hand. This way it is able to pick up on all the patterns there and apply this to anything that you might be writing in here. And while this worked, and I used this all the time over the past few months, this new feature works even better. Not just because it's more convenient and it's on the homepage, but also it picks up on the style in a better way. This is probably a mix of custom prompting on the background on how the style is implemented. And my guess would be that they don't just include it inside of the prompt, just like you do with custom instructions. Maybe it doesn't happen at the prompt level, but it happens at the user level. But this is just a guess. What I'm trying to say here is that it behaves a bit differently and better and I really like the implementation here. Okay, so in practice you can just switch over to your brand new style. Say write me an email to my boss about the broken coffee machine and it will use all of the in depth examples that you provided in the process of creating the style to write a custom email in your voice. Look at that. This is a tone that really matches the way I like to write. And if you want to use this yourself, I would recommend setting it up in exactly the way that I just showed you by giving it five examples. Now if you don't have five examples, go ahead and take some time to pull together different emails or just write five copies of something that you wanted to imitate. If you want to use this for blog posts, give it five blog posts. And then you're going to go in here and say create and edit styles. I'm going to create a custom style and they actually change this on the day of the release. This was its custom instructions, but that's a bit harder to navigate. And giving it examples always did work better for the styling. The only thing that works better than giving examples is actually fine tuning the model, but that's a whole different conversation. So what you're going to do is click add writing examples and then you can either upload or paste them like I did. So just paste this in here and really all you need is just email one colon, then you paste the email and you do the same for email. 2, 3, 4 and 5. 5 really does work the best. And once you say create style, it will analyze all of it, give it a descriptive name and you can start using this in your future writing. Afterwards you could go in and say edit style manually. And this is the cool part, it actually used proper markdown formatting to delimit the various examples you gave it and it wrote up a paragraph that exactly describes your style. And if you want, you could now take this and bring this into another LLM like ChatGPT. You could liter copy paste all of these examples go into chat.OpenAI.com or chats.com I guess. And now you could say customize ChatGPT and paste this inside of the second part of your custom instructions. Now, the problem with that is it might be a little long. I guess you can remove one example which will not help the quality here at all. And now it will write in your style too. If you're curious about the difference in tone, we'll put up both examples on the screen right now. You could pause for a second and check out the difference between the same style being applaud and Claude. And with the new ChatGPT model that sounds more human and is better creative writing, it's really just personal preference what you want to go with. I would say that once you create your style, you should run this for both and make your decision on what you want to use for writing assistance moving forward. And then to round this out, I just want to point out that Claude with this update included something that I've been asking for from ChatGPT ever since they released custom instructions give us different presets. I have so many different styles I like to use, so many different presets, so many different pieces of context that I like to include in the custom instructions. And every time I have to go through this process of copy pasting it into it here. Whereas in cloth you can just switch easily here, if you want to context switch, you can easily use the products to do that. So this one we use to write descriptions of community events. I could easily apply the formatting of the descriptions and pick my tech mentor style that is typical for my emails. Done. No copy pasting. And yeah, sure I could download some extensions, but I have multiple devices, I have the desktop set up, I have my laptop for travel or working from coffee shops. And I like to use my phone. And I want these presets every time, not most of the time. This does that. I really like it. Go give it a shot yourself. And now let's move on to the next piece of AI news that you can use. Okay, so next up, I wanted to show you this platform called seowriting AI and they're the sponsor of today's video. But I'm so glad they reached out because what they offer is something I'd be looking for help with and that is creating high quality articles. And this is one of the earliest AI use cases. But the interface and the features they included here really sets us apart from just simply prompting inside of ChatGPT, one of their standout features is their one click blog post tool. And I'll show it to you now. But essentially this is the easiest way to create a blog post, including images, videos and even citations, all you need to do is pick a topic, make a few selections based on your preferences, and voila, you've got a complete article ready to publish that goes beyond what would be possible with just an LLM. Let me show you what that looks like in practice. So once you log into the platform, all you need to do is click the one click blog post creator. And in here we pick our main keywords and titles and that's the bulk of it. Everything else is customization related. So with the title, I'll just go with a guide to Kitesurfing in Dental for Brazil, just because I came back from my trip recently and I absolutely loved it. And the main keyword will be Kitesurfing Brazil. You could even generate the title from that. And at this point I could just run it and it would generate the article. But I want to show you a few options down here. So beyond some of the obvious things like the article size and the model selection, that makes this more expensive or cheaper to run, you could also humanize the text, which makes it sound less like an AI wrote it. And there's a whole lot more in here that you can do. But the two things that I really wanted to highlight is once that you have this media hub where it generates images and includes YouTube videos as a part of the article for you, default settings here work just fine. And the second interesting thing here, in my opinion is actually this connect to web area and you can give it access and it will analyze hundreds of the most relevant websites with statistical data and reference those websites and articles. In this one, you could even turn on inline citations so that when it references something, it includes a little citation mark and you could even turn on for it to have the URLs at the bottom. So with just these changes, I'll say run and there you go, your article is ready to go. And if you want to create multiple of these, the SEO writing platform even has a bulk article generator right here. On one last thing, if you have a WordPress blog, you could link this and it would just auto post these. No further steps required. So if you want a custom AI system that bulk generates articles for you and automatically posts them, check out SEO writing today. Simply click the link at the top of the description and use the coupon code AIADV25 to get 25% of a premium plan. And now let's move on to the next piece of AI news that you can use. Okay, next up we have the chatgpt app adding a new feature. For now, this is Mac only, but essentially, as predicted, they're adding more applications that connect with the app. At this point, they're routing out the different development applications, most notably the VS code forks like Cursor and Windsurf. On Twitter, they even confirmed that they're excited to bring this beyond coding application soon, which I think is fantastic because ChatGPT could benefit from a lot of the context that happens in my everyday work. And I think it's particularly interesting that it works both with Cursor and Windsurf now, because you can use features like GPTs with these apps. Although that might not be too meaningful because Cursor in itself is a better version of a GPT, but baked into the code editor. But nevertheless, you could do things like have code sitting in multiple applications and then ChatGPT pulling it all together and you can prompt on top of all of it. Now, no more barriers of copy pasting between the different apps. I'm excited to see them extend this. But the one feature that everybody really wants is the access to the camera on the phone and to your entire desktop. So give it access to all apps, not just the development apps. And as you can see from the communications, this is just a question of time. And eventually these improvements will arrive on Windows too, extending your AI assistant beyond the boundaries of the current box that it lives in. Okay, next up, this could easily be considered the biggest story of the week, and that is the leak of OpenAI's Sora. You might have already heard this, but essentially some artists that got early access to Sora to participate in the testing decided to go rogue and publish their access to Sora on the Internet. Now, this whole thing was taken down within three hours, but a lot of people managed to generate some videos in the time window. And now these videos are freely floating all across the Internet. Friend of the channel. Theoretically, Tim pulled all of these together and uploaded a YouTube video on his own, where he compiled all of the 85 videos that people generated with Sora here. This includes all the most significant ones that I've seen go viral into one YouTube video. So if you want a glimpse of the future and probably the state of the art video generator, then you might want to put on this video. And one thing that I did is just turn off the audio here and I put on the Blade Runner soundtracks in the background and kind of had a glimpse of the future this way. Now, I'll say two things. First of all, I don't think leaking things like this is the correct way to handle this. At the end of the day, a sort of rebellious act like this only benefits the anti open source people as this can brought up as a great example of why things should not be publicly available and why they should not trust artists or other third parties with these tools ahead of release. I understand the frustration from the artistic community, but I do not think this is the way to handle that as it only sets a negative precedent for these sorts of things. And let's be real, the leakage of Sora is not a story that is going to be reported on worldwide. This is something that is talked about on channels like this and at the end of the day I think this is a net negative for the open source movement, which I personally am a big fan of. Anyway, this thing has been leaked and we have some examples. So let me leave you with a few comments on what I saw here in relation to all the other video generators that we test every time they come out. I think the big headline here is all the promo material was heavily cherry picked. While some of these examples here do still beat some of the state of the art video generators that are out there right now, from what I can see, the consistency of this thing is really all over the place, more so than some of the other generators out there. Now maybe this was not the final version, who knows. But compared to the multiple video generators that we have right now, some of these clips are just not that good. But at the end of the day, I got to say that one thing that blew everyone away about the initial demos was the fact that the hyper realism is truly next level. And all of these demos and leaks actually do prove that Some of these hyper realistic shots are indistinguishable, but the spaghetti benchmark is something that is yet to be solved. The quality overall, probably state of the art, but a bit underwhelming. I think people thought that they were further ahead than this leak showed. And with that let's talk about the next piece of AI use that you can actually use. And that would be Luma Labs Dream Machine being completely relaunched. And look, this doesn't happen often. A brand new application coming out and being completely rebranded and relaunched just a few months after the initial release. This in practice only means one thing. They did not see the usage numbers that they wanted, so they went with a complete overhaul to speak to a new user base. And they took a very interesting approach, quite different from what we've seen from all the other video generators up until now. They're tailoring their entire user interface towards a few specific use cases that seem to really make sense in their opinion. Now, the main problem of all video generators these days is that they're fascinating. On the first look, you really get these wow moments. You can generate everything. Look at that. A dog with sunglasses riding a skateboard. Wow. But then quite clearly, you realize that it does lack practical use cases. Sure, there's a few that we point out. Certain people create these short films or mock ads that work and are interesting as an example of what's possible with AI video today. And I've seen interesting and creative uses of its ability to transition one shot to another. But correct me if I'm wrong, the main use that I've seen in the mainstream of this so far is in music videos or in abstract visuals, where it is accepted that it might be a little trippy, a little different, and that is considered a good thing. Now, Dreaming Machine and its rebrands did a few things. First of all, they added these boards right here where it focuses on a few features. So one of the features is actually referencing a specific character for character consistency. Now, this is one of the big things that is missing in all these video generators. And with this feature, they're shooting their shot at that. If you could do consistent characters, you could do narrative stories that actually make sense to the viewer. But look, even within these four pictures, these are different styles. And I'm not sure I would say that this is the same person as this right here. But now I could go ahead and use one of the presets here and create more images like this and then turn them into videos. So they added this intermediate step of image generation before it actually creates the video. Now, obviously, if you create the video, it's going to introduce more variables and therefore more inconsistency. But as you can see, even based on that one character in the same style, this don't look like the the same person. Right. The jacket is different every time. The structure of his hair is different every time. Even the color and density of his beard changes slightly. So I would say nice try, but this doesn't really work. On top of that, they added this Brainstorm feature where you can use it as a creative companion to brainstorm new ideas for visuals. I think this is sort of interesting, but not sure how sticky this is. So there's a few more things in here, like mixing a style with a specific image. And generally speaking, the focus of this update is avoiding prompting. It wants to press buttons instead of having to type up text. That requires more knowledge and effort. At the end of the day, I'm not exactly sure how useful this has become. I think the.