Serious ChatGPT Upgrades & More AI Use Cases

Oh, wow. Let me tell you, this week in AI has been serious. We had releases out of OpenAI like we haven't seen them in months. Heck, maybe this was the busiest week of 2024. Yay. A new update to AI video generator had its viral moment all over social media. And in this video I'll show you how to do that for free. And so many more stories. So without further ado, let's dive right into all the AI news that you can use that came out over the course of the last seven days. Trust me, there's a lot here. ChatGPT actually released a brand new interface feature. It's been a while since we got one of these and this one is coding focus. As seem to be most OpenAI announcements this week with the Dev Day. But I want to show you what they did here because they did sort of a copy of artifacts. Not really. This is not set up to run code. It's more of an ideal interface to edit code inside of ChatGPT. Let me just show you. I think that's going to be the easiest. What we got to do first though is we got to navigate over to this new model GPT4.0 with Canvas, which is in beta right now. And let's start by looking at code, but you can also use this for writing, as I'll show you in a second. And this simple little prompt generate me a simple Selenium script that opens ChatGPT is going to be just right for our little demo here. And if you're not familiar, what I'm asking for here is a script that operates the browser for me where it goes around and presses buttons, types things in. And it did that and it automatically opened up the brand new Canvas. And as you can see, this is not just a preview, this is also editors. So I could remove letters, add new lines, write new code. And we have a few really useful features here that help with editing and improving the code. First things first, you can select certain text and then ask ChatGPT just on top of it. We had this feature before. You might be aware of the fact that you can select things here and then quote them and talk to them like so this is sort of the same thing, but it doesn't end there because here at the bottom right, a bunch of new buttons were added. Plus we can always navigate back. So if we make a step that we don't like, just like inside of a code editor or Word or something, you can always take a step back or forward, but back to this button on the bottom Right, because this hides a lot of useful functions. Let's look at these. Add comments, add logs, figs, bugs, port to a language and code review. These are all coding specific as that's what we're doing right here. But they're quite quickly explained. If I say add comments, it will go through and it will add comments to every single line of code. This is especially useful for beginners who do not know what the code does. Comments are just natural language that don't do anything in the code. Explain to you what is going on here. So you can see, this step opens up the ChatGPT page by navigating to the URL. All right, but it doesn't end there. We could go in and add logs. What this does is add different print commands so that as the program runs inside of a code editor, every time something happens, it gives us text. So for example here, after it goes to the website, it actually says ChatGPT page loaded successfully. Bug fixing is pretty straightforward. If something doesn't work, it goes ahead and tries to figure out how to fix it. So if you encounter errors, this is a quick way to fix them. And then the last two options are quite simple. Portal language is just translating this to a different coding language. So with this interface you can go to, for example, C like so. And all of these things could be done with prompts before, but you had to manually type them in and then hit Enter and you couldn't even edit the results. Now you can. So this is way more user friendly. And last but certainly not least, you have this powerful function of just reviewing the code. It just looks at the entire thing and starts making recommendations on what could be improved. And this is where it starts getting weird because we're getting closer and closer to this agentic future where ChatGPT will just do all of this on its own. Right now we're in this weird middle ground. Let me show you what I mean. So, Obviously we prompted ChatGPT to write the code. Great. Now we click the button for it to review its own code and we have all of these comments here on the right saying using system commands can be unsafe and lead to security vulnerabilities. Okay, I just say apply and it will prompt itself again and fix that by itself. And then there's a few more comments that I could apply, like so. And as you can see, it wrote the code, it came up with how to improve the code, and I'm just sitting here and pressing a button, telling it to go on and make the whole thing better. At this Point. It's not hard to imagine a future where it does all of this automatically. Matter of fact, these apps are already coming out. If you check out the last upload on the channel, it's a new app that just does all the coding for you. You just talk to it like you would to a human, and then a bunch of agents do the work in the background. And these features are available in ChatGPT now tool. Now let's have a brief look at the writing use case too. Okay, so I'm just going to go with the good old write me an essay about penguins. All right, let's see. And there we go. Code editor. The canvas opened up and now I can edit my text in here. This now acts more as a word processing document. So I could make changes here, remove some of these ChatGPT type words, make local style related decisions, I guess. Or I could just ask ChatGPT to make this more concise. Let me tell you, this is so much better than having to go over to Microsoft Word and then copy, paste back and forth, back and forth, piece things together. You could just work in here. This is great. And then as you can see, these options down here change again. So I can add emojis. Who would want to do that? Usually you just want to remove emojis. You can add a final polish or a bunch of other options. And as you can see, these are all things that were hidden behind prompts before this. I had to create free notion templates with all of them and create videos to show you what was possible. But increasingly these are being brought into the interface, which is just great to see. So final polish apparently improves the structure and the flow and improves the clarity and the grammar. And this would be my tip. Whenever you do something here and you want to understand what's going on under the hood and what it's prompting, well, you can always see over here. Now let's try another one. Let's change the reading level. And these were some of the first prompts that I shared. You know, something like write this at the level of a five year old. You can just prompt it with this little slider. Now there you go. Not all penguins live in icy places. Another one obviously is the length. You can make it super long or super short. Or again, you kind of just have this global option of saying suggest edits and it will figure out what could be changed. And then you can act upon those recommendations or choose to ignore them. Okay, sounds good. And it's done. Let me tell you, this is such a blessing. Compared to copy pasting into other apps. A lot of other applications implemented some of these features. I mean heck, even Apple Intelligence is doing it. The notion you have instead of all the Microsoft apps, you have them with Copilot now finally you have them right in here. This is really a welcome addition, especially if you have the ability to navigate back and try things, break things and you can always restore. You know that you have this non destructive workflow and all these handy little prompts and buttons. Now OpenAI, if anybody's watching this, please for the love of God add some toggle in the menu to have an advanced button or ideally even free that I can customize with my very own prompts. Please let me do something like a custom instructions preset where it can rewrite it in my style. Just one button. It's all I'm asking for. But I've been asking for so long at this point I lost all hope. They ask you how you are, you.

Just have to say that you're fine when you're not really fine.

Thank you for this feature. Now let's look at some other new releases from OpenAI that came out this week. All right, so first up, you might have heard that OpenAI's Dev Day happened this week and this year they took a different approach. There was no live stream, just a bunch of invited guests and they presented a very, very developer focused update here. And I figured that giving you a brief overview of what came out and how to use some of these things might be a useful thing because most of the things they talked about in this event are not in this announcement category. Things that be coming one day. No, they're available today and there's actually different websites where you can use them, try them and then build applications on top of that. And in case you're just a consumer and not planning to build with all this, it's good to see because you can expect this being built into all types of applications. Just like there's an AI text writing feature in pretty much every app these days. So first up they released the Real Time API and you can think of this as the programmatic version of the advanced voice assistant, giving developers to essentially built in advanced voice features into their apps. Here's a little demo from a video that OpenAI for developers uploaded recently.

Donde esta el bano? The word banyo is pronounced ba nyo with a soft nye sound like in the English word canyon. Can you try saying that again? Donde esta al bano?

Perfecto, Amazing, right? You can only Imagine how many applications would benefit from a real time voice interface that has some level of intelligence in the background. I mean, first thing that comes to mind for me are these annoying hotlines that all the different banks and insurance companies have in place where the robot has a fixed path laid out that it needs to follow and often it doesn't even understand what you're saying properly. Well, all of that is about to get a whole lot better soon. And there's actually a website that I found where you can try this now, because inside of the OpenAI playground, there's no great interface to do this. And setting this up for yourself takes a little more effort than most people would be willing to put put in. So in this site that I found on Twitter, it's called Live Kit, and apparently they provide their own API keys, which, by the way, this thing is pricey. When it outputs speech, it costs 6 cents per minute, and when a user is inputting speech, it's 24 cents per minute. Now, OpenAI has a history of making this cheaper over time. But over here, at the time of this recording, October 3rd. Okay, editing Igor here, at the time of this recording it was free. But by now, Friday, October 4th, you need to link your own API key here that you can get inside of your OpenAI playground. But OpenAI inside of the playground enabled a real time space too. Now, so you can just start the session here and that's probably the best way to do it. Just do it inside of the playground like so. And you can start talking to the voice assistant. There you go. Okay, testing the voice assistant.

Hello, how can I help you today?

There you go. Life Kit provides you with this website without logging in. I can kind of just go in, say connect, and once I give it access to my microphone.

Hello there. How's your day going so far? I'm here to help with anything you need, so just let me know what's on your mind.

All right, that's good to hear. I'm recording a news you can use episode. What do you think of that?

That's awesome. A new music episode. Sounds like a lot of fun.

No, no, no, no, no, hold up. It's not a music episode, it's a news you can use episodes.

Ah, got it. A news you can use episode. Sounds like it'll be packed with useful and interesting information.

Oh, indeed. I think you get the point here. You can put in various instructions here and try this out. As I mentioned, very similar to the advanced voice mode, but up here you actually have a few use case Demos, which, hey, use cases. That's what we always like to see. By the way, shout out to the guy who last week commented that, hey, if you had a drink every time this guy says use cases, you'd be in the hospital right now. Yeah, you bet. That's what this channel is all about right now. But as you can see, you can set it up to be a Spanish tutor, opera singer. Oh no. Or a Valley girl. Let's have a listen to this one.

Hey there. What's up? How's your day going?

Yeah, my day's fine. How's your day?

Oh, that's awesome. My day's been pretty chill so far, you know, just enjoying the good vibes.

Oh, I bet you're enjoying the good vibes. All right, enough of this. That's all crazy good, right? It's okay. So there you go. Here you could be testing this, but that's a lot more came out of OpenAI. Most notably the ability to fine tune their vision models, prompt caching, and a bunch of other developer features. Let's start with prompt caching, because that's really easy to cover. We talked about this a few weeks ago when Anthropic released it, and before that Google Gemini actually releases first. It's essentially a mechanism to save money. If you rerun the same prompt a lot of times, if certain tokens are repeated over and over again, you can automatically expect a 50% price reduction in running the API. That's basically what this does. What is nice about the OpenAI one is that it's actually automatically applied and it even works for O1 mini and O1 previews. So basically a massive cost reduction on a bunch of their APIs again. Now let's talk about fine tuning for the Vision API. Because we played around with fine tuning a lot with the AI Advantage team and we developed several step by step guides in our AI Advantage community, with more on the way. Actually, we're doing an entire learning path just about fine tuning, because I think there's so much value hidden there, particularly the use case of fine tuning. A model on your exact style is something that goes further than anything you can do with prompting. Like heck, if you fine tune something like GPT4 or mini properly, it starts swearing, good luck doing that within ChatGPT. And now they're bringing the ability to fine tune to their vision models, which is interesting, but I'm not exactly sure where this will find its applications. But time will show. I certainly think this is powerful. They pointed out potential use cases in E commerce stores where you can train the Vision model on the particular items that are being sold and then it can be used to make better recommendations to customers, or it could even process refunds and things like that. I think everything where the Vision API wasn't quite good enough up until now, this is going to open up a new level of quality that you can achieve, a new level of precision too, because you're just going to give it a thousand images of different versions of your damaged product and then it will be able to identify the thousand and first image in a more reliable manner, rather than the general purpose model that is the general vision API today. Or you upload 50,000 images to train the model because that's the maximum it would take. Now, me and the team had a look at this and it's actually a bit more technical and intricate than just fine tuning with text. So we'll need a bit more time to wrap our heads around this and really put this to work in a meaningful way. But my suspicion is that this one will open up a whole, whole lot of possibilities that the Vision API just wasn't good enough for up until now. And just like the other announcements, you can start fine tuning Vision models today. They're also offering this new feature called Model Distillation, which will allow you to use one of the big models like O Preview to generate data for one of the small models. So you can fine tune it for specific use case. So as they show in their example, if you have all of these customer complaints and you need to categorize them for your customer support agents to then handle well, on the one side you could input all the categories of complaints that you have and then the LLM would go ahead and generate various customer complaints and what they might sound like. If you have enough of these, you can fine tune a little GPT4O mini model which is really cheap to run and will be very, very reliable for this one use case of categorizing complaints. I hope that makes sense. And there's a few more things. I don't think any of these are super relevant. One of them is that there's a new Whisper model and this was slightly confusing in V3 Turbo, V3 Turbo Large and V3 Large actually released a year ago in November 23rd, but now we get this Turbo version that is slightly worse but way faster. You can also try this out for free on this hugging face space. This transcription is supposed to be super, super fast. Let's have a look at how this performs in practice. Yeah, there you go. As you can see the latency on that is super low. I think this week marks the end of an era where we had latency with various text to speech models or AI assist assistance. Now, between this, the real time API and advanced Voice mode in ChatGPT, it all happens at the speed of pretty much talking to a human, doesn't it? And then lastly, there's this one feature if you go to the playground and you go to the chat interface where you can try all the various models and fine tunes if you have them down here. There's a new feature here which automatically generates system instructions. It creates your system prompt for you, so you can just click this. By the way, I think this is the first time OpenAI is embracing this, this universal AI logo of these two free little stars. So I can say something like, I'm translating English to Spanish now to generate our system instructions for us, which is interesting. I'd love to see a feature like this inside of ChatGPT where you could just auto generate custom instructions. I mean, right now we had to craft custom prompts and teach it to you on this channel in the community, but you could just press a button and everything could be happening in the background. Again, as I say so often, with a lot of the stuff we teach, we're just a few weeks or months ahead of OpenAI and then they catch up and do it in a more intuitive way. But nevertheless, I think there's value in, in doing these things manually because you get ahead of the curve a little bit and maybe it might open up an opportunity that others are not aware of yet, because it takes a little effort. So as you can see, many of the prompt engineering techniques that we teach, markdown formatting, delimiters, a two shot prompt in the end, yeah, this is really solid. And now you can generate these instructions automatically inside of ChatGPT. Just as a reminder, if you want to do this to your prompts, we have this free GPT out there that we call Sam, the Prompt Creator, which allows you to do the same thing with prompts. It really fleshes them out and does all the formatting and prompt engineering for you. Just as a quick recap, in case you haven't seen our videos on this, this is best used if you're going to be reusing the prompt regularly and you want consistent outputs, or you're going to be putting this inside of a pipeline or automation where you won't be checking the outputs. That's where maximum consistency and precision are essential. And those are the cases where going the extra mile with prompt engineering. And using a tool like this that takes a simple process prompt like this first line of the instructions here and turns it into something even more intricate with literally the click of a button is probably worth your time. And look at how similar that looks to what OpenAI is doing here. I feel flattered that we released this months ago and for free. Okay, next up, we have a brand new update to AI Video model. And hold up before you skip this because you don't care about creative things, this one actually caused a lot of reactions because it has a really fun feature in it that you should probably have a look at. It has these presets and you can even apply this for free. That's why we made it App of the Week in our weekly newsletter, because you could just go in here and pick one of these pika effects, okay, so you could inflate something, you could melt it, explode it, squish it, crush it, or cakeify it. And now we're at an interesting point of the video because I can say something magical like, hello, video editing team, hello there. Let's have some fun with this. How about I stand up and you apply some of these effects to me? One more. All right, I think you get the point here. It's really fun. And you can do this on a free plan. Right now, just upload images, fun little capability. And now let's move on to the next one, which is less of a piece of news that you can use, but more like, hey, all of this stuff that we're looking at actually has some real world applications and people are hiring for it. Namely Mr. Beast put out a job listing for AI concept artists. My guess would be that he's helping with thumbnails or maybe even set designs. I think this is really interesting because he clearly sees the advantage of having AI as a part of your workflow. The speed of iteration that you get is simply unmatched with any other workflow. And on fast paced productions like Mr. Beast Does Them, this does make a lot of sense. And even though he has faced a lot of controversy, I think it's undebated that the business empire that he built is significant. The fact that he's looking for people with these skills, to me only indicates that this is sort of the beginning of a trend. I think we're only going to see more of this, and people who are skilled with these tools will be demanded more and more as the market starts to understand the advantages of having some of these skills that popped up in the recent months or maybe Years at best. Just thought this was interesting enough to highlight and now let's move on to the next one. Alright, this week for our sponsored segment, I actually want to give you a brief look inside of our community because it has evolved so much over the past months. I know I keep referencing it, but that's because a lot of the content here on the channel gets inspired by the community. Now we actually did a thing where we restructured the whole thing into three different districts. We have one for LLM, one for Creative, and one for no code applications. And for example, in the LLM district we open up a brand new space which acts like a private forum just for advanced voice mode and people have been sharing what they're using it for. There's vivid discussions here and look, check this out. This is my favorite part. If people come in here, the introduction area is so vivid. Look at that. 10 to 30 comments on every single thing that has popped up more than a day ago. And for people looking to learn practical stuff skills, we develop these learning paths that keep getting released. As I mentioned earlier, we're working on one for fine tuning and also bot building right now. But as of right now you have the full and updated prompt engineering course in here, plus the GPT building learning path and the advanced LLM prompt engineering path that is nine hours long. That's a lot of advanced lectures and they're all presented by me. So for anybody who doesn't just want to keep up but actually get ahead of the curve by learning some of these technological skills related to generative AI, well this community is my best possible attempt at solving that. It's really the ultimate way to master these AI skills along with personalized support to help you every step along the way. You can check out more details in the link in the description. If you have any questions, simply leave a comment or shoot me an email on IGordvantage.com all right, now let's get back to the next piece of AI news that you can use. Okay then we have some serious news. Out of Notion AI because they didn't just change the logo of their AI offering, but they also added a bunch of feature and one of them looks like something I might be actually using all the time. Oh by the way, if you're new to the channel, I kind of use Notion to organize my entire private life and all of my business activities as I like to have everything in one place. So this will come in very handy if I just open up a random page over here. This is my idea. Quick capture Page whenever lightning strikes. And I have a genius idea, go in here and quickly capture it. Let's just open up one of these pages to see what we're dealing with here. Incredible names. Guac Madocious. Okay, I'm not sure what was going on in my life when I captured this, but now I can get a feeling for the type of of quality we're dealing with here. But the point is this. There's a new button here on the bottom right. This is their new AI Assistant logo. And beyond the basic functionality, you now have extended features like attaching files, which is all right, but let's be real. This is Notion. All my files are probably already in there. And this is the one feature that I was referring to. Generate and edit docs in your own style. So if I open these random copy ideas that I had here, for example, for the community AI, for people who have a life and want to enhance it, that's sort of fair. I could actually go in here and say, ask AI. And in here I could do custom prompts like rewrite this with. And then if I type at, I could pick any page inside of my notion. So if I have a style guide or I want to combine this with something like the incredible name that we just looked at before, I could say combine this with the content of and then pick the other document and then it will merge those two and look, it figures out a way to combine those. Now, does the sentence make any sense? I don't know. Probably not. But you get the point here. This is super powerful. If you have your business built in here, you could easily transfer numbers, essays, style guides, design manuals, or even event meanings into other documents. You could summarize, combine, rewrite in the style of all the tricks that we have within ChatGPT, you could be using in here now. Heck, it would probably make sense to create several pages that do just that. And luckily I already have a lot of that set up. So, for example, here is my educational Persona that tends to resonate with our educational type content. I did the same thing for the other Personas we might want to target, and now I have a great way of using it with other materials. Whenever I'm crafting something marketing related, I can just add, mention this Persona and craft it specifically for them. This is quite amazing. And I'll start using this concrete feature all the time now. And if you work inside of Notion, you might want to do it too. All right, next up, Cling released a new feature, Lip Sync. So what me and the team did Is we ran a bunch of images through Cling and some of its competitors, like Heygen and Runway Lip syncing. And here are the results for you to compare. Do you think you know me? You have no idea what I've been through. Every step I take, every decision I make, the very first clip within Cling is decent. Do you think you know me? You have no idea what I've been through. Same thing in Runway. Didn't even move the lips. Do you think you know me? You have no idea what I've been through. And in Haejin, half the mouth is sort of closed, plus the background does move. It's just a face. Okay, let's have a look at this cook inside of Cling. All right, let me tell you about this new dish. We're talking about seared scallops, golden and crisp on the outside. Yeah, I don't know about that one. That was a bit weird. All right, let me tell you about this new dish. Again, not moving the lips. Okay, next one. Hey, Jen. All right, let me tell you about this new dish. We're talking about seared. Okay, so this is not great. I think Kling is the best of the bunch, but that doesn't say much. Next clip.

You are stronger than you think. Every challenge you've faced, every setback you've endured, those were lessons.

Not bad, not terrible.

You are stronger than you think. Every challenge you've faced, every setback you've endured.

Here, Runway is sort of decent. And then. Hey, Jen.

You are stronger than you think. Every challenge you've faced, every setback you've endured.

Yeah, I actually really like Kling here, as it also moves the camera and stuff. It's not perfect, but it's the best of the bunch. One last comparison. Look at that. That's pretty good. For a fun use case like this, it might be good. Yeah, I guess Runway is okay. Okay. The static nature of hey Jin just isn't it? So I would say, out of the bunch, Klingk performs the best. So in hey Jin, you can include Voice generations from 11 Labs, which is a big plus because they have the best voices. But Kling can only do 10 seconds at a time. But if you can fit that in, I guess it's the best lip syncing tool in the market right now. And there you go. That's it for today. The guy that was planning to have a shot every single time I use the word use cases must be absolutely devastated and is probably at the hospital at this point. But for all you other wonderful viewers, I hope you enjoyed this, and I'll see you next week.