ChatGPT Mentions and 8 More Tools Are Here!

This has been an exceptional week for AI use cases, and I'm here to show you everything that you can put to work today. Because me and my team have been digging through the Internet to find the little gems, like this app where you can use your voice to practice interviews. And then we're also talking about the big announcements like code LLAMA70B or Midjourney V6DG. So sit back and relax because this week has been packed with practical applications for AI. First things first, let's start by talking about ChatGPT mentions. This got many people excited. So let me first summarize what this is about and then give you my take on why this is not as big of a deal as you think. So, so if you're familiar with the concept of a GPT, it's actually quite simple because you can just go into any GPT now and then mention another and then you can have both of them in one conversation. So if you follow the content, you will have recognized that when GPTs released, I basically said this is no revolution, this is nothing new, it's just a repackaging of something that we had in a way, way, way more user friendly way. And essentially the same thing goes here, because what you can do with these mentions you could have done before too. So if this GPT is supposed to talk to Innovator GPT, both these are based on instructions. And I could be just sending all these instructions to a new ChatGPT thread in here and then following up with the instructions of the Innovator. And that way I could have the two of them interface. But now it's way easier. So beyond what it does, let's talk about how to use it and how not to use it. So one thing that I notice is that it trips up with many GPTs that already exist, especially the ones that have a specific communication sequence. So what I teach when talking about GPTs is that you want a communication sequence in the beginning, a way of interacting with the user. Here's step one, here's step two, here's step three. That's a really good idea if it's important, interacting with a user. If it's interacting with another GPT, it trips it up because the other GPT doesn't behave like a user. And often these GPTs are purposely built to interact with a user. So for anybody creating GPTs, watch out, because now with the Mentions feature, things change slightly. And for all ChatGPT plus users with this feature, just be aware that you can now use multiple GPTs in one conversation and the number one thing you will want from them is probably the actions part. So the ones that are really, well, combinable are the ones that do something specific. As an example here, this diagrams GPT has an action that will create a visual diagram for you. And that's why currently I'm experimenting with different ones that I pin to my sidebar and then just bring them into my conversation. Whenever I'm going to create a diagram, no need to go to a new conversation. So again, the best ones to combine are the ones that do one specific thing, ideally with an external facing action, because everything else you could have done with instructions before and if that was useful, people probably did it in some way and you can find it in the store. One last comment on GPT so as I've talked about many times before, you can think of ChatGPT as your assistant or your employee. And in this case when you mention another employee, you're kind of letting two employees work together, right? And using that analogy, in the real world that can go very well, right? Giving two people a task, they'll often figure it out. But why don't we have the ability to mention like 5 GPTs? Well, if you've ever given a task to a group of people, you will know that that is utter chaos. You really need somebody to take control there and to manage those five people. And therefore that situation is a completely different one than letting two people, aka two GPTs work together. That is quite simple. But as soon as you enter the realm of three or more GPTs, you really need a manager. And that takes a whole different approach than the product that we have right here. So as I've said in my AI transfer 2024 video AI agents are the direction that I see this entire product develop in. And these mentions are baby step on the track to getting a team of agents that work together autonomously. And the feature does make using ChatGPT a little better. So I'm glad they did it. Now let's move on to the next use case. So next up we have more new stuff from OpenAI, but this is targeted at developers. So we're getting two new embedding models, a new and slightly improved version of GPT4 Turbo. Apparently it's less lazy when it comes to code generation. They slashed GPT 3.45 API prices again and then updated moderation model. So all around this is a quality of life upgrade and cost reduction for developers using the API. This doesn't apply to you if you just use chatgpt, which is chat.OpenAI.com, but a few things I wanted to highlight here. First of all, this is the first time they actually publicly admitted that the GPT4 model was getting sort of lazy in certain cases. A lot of people talked about this, but they never really confirmed. So here you can see that the new version of the model reduces cases of laziness. And another thing that I personally am very much looking forward to is the GPT4 Turbo with vision. So this is where you're able to prompt on top of images, but right now the API is just super slow. It's just not reasonable to use it for most use cases. That would be interesting. Okay, so next up, let's talk about Bard, which has been getting silent updates under the hood without any big announcements. So during the last announcement of their new models, including Gemini Ultra, they had this little demo that people didn't receive so well, understandably. So the whole thing was edited and wasn't representing the actual model and how it works. But they also also shipped Gemini Pro, which was instantly accessible through Bard. But the thing is, Bard was just never that good. Usually people avoided it, but they have been upgrading it, they have been adding plugins here and performance upgrades apparently. And now the whole thing is actually fully up to date, if you didn't know. So the knowledge cutoff is not some arbitrary date in 2021 or April 2023 like GPT4, but it's today. And if you check out the Chatbot arena leaderboard, which is a fantastic way to track the performance of some of these different models. And if you didn't know, Chatbot arena actually uses the ELO system, which is used for ranking people in chess and also in many computer games these days. And essentially they have different benchmarks where they let two different language models battle it out and the better result gets the win and gathers points to rise in these rankings and in this system, in this dual base system, Bard actually ranks as number two. And it's quite close to GPT4 Turbo, honestly. Now here's the thing. This up to date knowledge cutoff might be one of the reasons for that, because it just has a lot of up to date information. But that's a big strength that you should probably be aware of because if you're looking to prompt on something that is, you know, happening right now, the browsing feature in GPT4 is absolutely terrible, let's just be honest. So, yeah, Bard is way better now and I hear from some team members that they actually like to use it for Research. So I added it back to my toolbar and I'll be revisiting it for some of my daily use cases over the next week. And let's see how that goes. But yeah, there you go. Apparently Bard is back in the game. Okay, editing Igor here with a quick update. Google actually, just before we uploaded this video, came out with Imagen 2 that they added into Bard. So now you also have an image generator inside of Bard that is actually quite good. So the images that I've seen so far from it are solids. Nothing insane, but not bad either. But there's an interesting fact here because they built in an AI watermarking feature into the images that you get here. They call this Synth id and the watermark is invisible. It's just embedded inside of the picture. And there you go. Now back to the video. All right, so the next tool I want to talk about is actually the sponsor of today's video. So let's have a Look at the 10 web AI website builder that allows you to create fully functional WordPress sites with just a single prompt. This is really good for portfolios. This is really good for sites that represent your business on the Internet. You won't be creating web applications with this. Rather than that, you want to focus on sites that convey information like a portfolio or a company website with all the details and links that are necessary. That's what this is really good for. And here I'll say I want to create an informational website. Next, what is my business type? So let's say startup colors structure in the company description. I'll say we provide automated data scraping for online businesses. I'll enhance this with AI too. So it fleshes it out a little more. That is great. And yeah, these pre filled services with AI work and that is it. It's creating the website. It's going to go through these steps and we have a first draft that is live on the Internet and then we could easily tweak it from here. And there you go. It created 10 subpages and we can now go ahead and edit all all the details in here. So as you can see, it doesn't take any technical skills here and once you're ready, you just say save and publish. And this thing is live on the Internet. Yep, works great. So there you go. Check out the link in the description to try out 10Web for free today. Thanks again for them sponsoring this video. And now let's move on to the next use case. And you know, before we get into another technical announcement, let's Actually come up for air for a second and let me show you this super fun one which was actually shown to me by nightclass from our community. So thank you so much for that, man. It's this really simple app that does a thing really well and that's just recreating GIFs with other faces in it using AI. And look, it's not perfect, but for quick comedic effect in a group chat you don't need perfect. So let's give this a quick try. All right? Simple little picture. Drag that onto my desktop. One minute later. Look at that. Very nice. Great success. So if you have an image of yourself on your desktop, this is fast enough for you to just reply to a message in a group chat with a custom meme like this if you so desire. Oh my God. I just love little apps like this. This is completely free. All it takes is one image. Have fun. And now let's talk about the more serious and actually significant code llama 70B release by meta. So I consider this one to be important even for the non technical people because this is actually a big deal. Matter of fact, it seems to be the best for code generation when it comes to open source models that you could actually run locally or fine tune yourself. And sure this could be measured in many ways, but one of them is actually letting it push program an entire game of Snake and most models are not capable of that. As pointed out by Matthew berman on his YouTube channel, which is more development focused. The only two open source models up until now which could write a full game of Snake were Mixtral and Mistral Medium. And now code llama70B is able to do that too. And I think that the best part here is really the fine tuning because what this allows you to do is you can take the model and fine tune it for your very specific use case. So here Rishabh on X went ahead and shared his results with a fine tuned model of llama 70B for SQL programming. Okay, if all of these technical terms don't mean much to you, don't worry about it. What you got to know is that on this graph this fine tuned model outperformed everything including GPT4, Turbo and GPT4. And that's because you can fine tune it for a specific use case and make it be specialized. This is not possible with GPT4 yet. This is a big one we're waiting for. But fine tuning is super powerful and you can do it on Llama. So for coders and company, this release is actually massive because it comes with A license that everybody can use with the exception of essentially Google and Amazon. So big release right here. You do have to apply for a form to get access to it, but usually Meta is super quick with their forms. Okay, so with this you really get the ability to run a local GitHub copilot. That is actually amazing. Okay, but now let's move on from stuff that is interesting for developers, for stuff that will be interesting to every single map user out there. And I ran into this app and I just found this genuinely amazing. They found a use case for GPT Vision, which the world needs, which I need. It's very simple and this is a Mac only problem. And the point is, every time you take a screenshot, like so, okay, Command Shift 4i screenshot this website, that screenshot gets stored on your desktop. Here it is. And its name is something super undescriptive. So all this app does is it allows you to plug in your own API and then it names the screenshots based on the content of them. Look at my desktop. Good lord. Just right now, there's about 60 screenshots there. So after downloading the app and entering my API key here. So I'll just go ahead and select some images on my desktop, click this icon at the top, rename selected, and there you go. And name them. Modern living room with green screen. Yep. Perfect. Blank billboards. New York City street. Yep. It even recognized the text in here. And also this one is correct. So I gotta say, this is one of those where I didn't know that I actually needed this, but I really did. I take a lot of screenshots and no, this thing is not free. The first five images are for free and then you have to pay. But honestly, if you take a lot of screenshots, especially furry work, this is a really simple yet very effective use case and a perfect example of how some of these things just create a new world inside of these computers where we won't want to go back to how it was before. And yes, it's probably one of those that over time will just be natively integrated. But right now, this is the way to do it. I'll be renaming all these screenshots and with that being said, let's move on to the next one. So I got two more really interesting use cases for you this week. One of them is behind the midjourney paywall, but it's really worth talking about because the IS key came out with a midjourney V6 Niji model, which is their anime model. And this Seems to be so capable and I personally absolutely love this style. Now, instead of just showing you random examples or prompting it myself, I will show you examples from one of my favorite artists on the Internet and a member of our team specializing in stable diffusion, Image Journey, Hades. And Hades is really good at prompting artistic things from these models. So look at some of the images he got on day one with NI V6. And we'll just put some appropriate music into the background and enjoy this for a second. Hold on, Wait a minute. I don't know if that's appropriate. Try again. Editing team. Okay, that's way better. Let's keep going. Amazing. So, yeah, incredibly powerful stuff. And the way you use it is just by saying slash settings and then you would change your settings here to NI Model vc and that is really the easiest way. Now anything that you prompt with slashmagic will have this new model applied. Okay, so last but definitely not least, a sweet little use case where you can actually use your microphone to practice for mock interviews. And I think this is such a fantastic use case for many people, especially younger people. Here you can practice your tech related interviews.

Hi, this is Steve. Can you hear me okay?

Yeah, Steve, I can hear you well, great.

We are conducting a mock system design interview. You'll have five minutes to to go through the problem. The problem statement is to design a blog app that supports creating and reading blog posts, each with its own set of blog posts. Does that all make sense to you?

Yeah, makes a lot of sense. Let's do it.

Great. Please draw your design on the whiteboard or talk me through it if that's easier for you. How does that sound?

Okay, let's do it. Ready to go. So as of now, this is just a freely accessible demo, but they will be building this out to various scenarios. And there you go. Those are all the things that you can use AI for that I found super interesting and useful from this week. If you want to learn more and haven't got the massive notion template with GPTs, Prompts, Styles, your own prompt database and so much more, then you can get that to signing up to our weekly newsletter link in the description below. Again, it's completely free and even includes prompt generators. And other than that, I'll see you next Friday where we'll be looking at the next batch of AI use cases that come out over the course of the next week. I'll see you there.