Best Ways to Use OpenAI o1 & More AI Use Cases

This has been a very interesting week in AI because it's been a while since we had such a large divide between opinions of people saying that the new releases are utterly useless and there's no innovation whatsoever, and then others saying that this is actually the most groundbreaking thing since the release of GPT3. Now what I'm talking about here is the release of O1 that happened last weekend. And this week we pulled together a few use cases and I'm going to show you a little workflow where I believe that O1 is actually preferable to any other model on the market. But it doesn't end there because this week we received a variety of AI news that you can actually use, ranging all the way from the O1 use case cases to Microsoft releasing a wave two of their core AI offering and state of the art AI avatars and so much more. We went ahead and researched and tested all of it for you and that's what we show you every week in an episode of AI News that you can actually use. Alright, so first up, let's talk about the developments in OpenAI's brand new Model 01, because now roughly a week has passed since the release and a lot of the hype that came of the release has settled down now and some real world use cases are popping up. And that's exactly what this show is about. We'll be looking at some of them. But before we do that, I just want to update you on the fact that the rate limits have been increased. Now you have 50 messages per day with O1 mini, not just per week. And with the 01 preview model, you get 50 messages per week instead of 30. Also I should note, if you want unlimited messages, there's one easy way to do that is if you have a paid account with poll. And now let's talk about what this can actually be used for in the real world. And I have several examples of what people have actually been using this for that's set themselves apart from what has been possible with any large language model so far, starting with a use case that was actually featured by OpenAI themselves, which is using OpenAI's O1 for legal work, concretely for drafting contracts based on templates. So to be clear, they use O1 not to write a contract from scratch, but to edit and improve an existing one. And as you'll see with some of these use cases that I have here, that is sort of the reoccurring theme. It's more editing rather than creating from the ground up. Seems to be the strength here. They also show some math use cases here. But rather than showing you some specific calculations that it performs well at, I wanted to show you this graph that displays the accuracy amongst multiplication. So if a number has, for example, 10 digits and it's multiplied with another number of 10 digits, well, only in 3.8% of the cases does O1 mini get this right. That's next to nothing, but it's a whole lot better than the sea of red of GPT4. And that's why I like this graph. It really clearly shows the improvement here on these numbers where GPT4O gets 23%. So that would be four digits times four digits. O1 gets it right 100% of the time. A real improvement in calculating and long term planning, isn't it? So how can we use that to our advantage? Well, it can build applications and this has been sort of the dominant use case all across the Internet here. In this case, it actually builds a fully functional chess game, including an AI opponent that plays against you, and it actually works. This is one of the tests that have not been possible with any other model so far. But still, my problem with this is most people are not going to be building chess games. Well, that's certainly impressive. It's not a practical, everyday use case. But hold up, that doesn't mean that you can't be taking advantage of this brand new model because I found another post which reflects my personal experiences with this model so far really, really well. So this is the context of building an application, but this absolutely applies to whatever else you might be doing with AI already. So listen closely. Dan MCA here is saying the following. O1 mini is the architect. Explain my requirements and have it create a detailed design document with step by step instructions for each module. So O Mini does the strategizing, the high level thinking, the step by step planning, and then he uses Sonnet 3.5 as the developer to actually write the code, to actually do the work. It generates the code based on architectural document produced by O1 Mini. And this, let me tell you, this is the correct workflow and this is the correct way to use it. Let 01 do the thinking, the strategizing, the brainstorming, let it be the mastermind and then take what it produces and bring that over to another model that might be better at a concrete thing, like for example right here. Sonnet 3.5 is way better at code generation than GPT4.0 or 01. But O1 shines at thinking. So we're at this interesting point in time. Where we're sort of in between two phases. We're not quite in this agentic feature that has been promised and that is most certainly coming, but we also don't fully have to rely on these chatbots that do nothing but assist you. And I'll end the segment on the point that people have been running O1 through IQ tests and it actually scores at 120. If you're not familiar, 100 is the average IQ of humans. This shifts over time as human performance on these tests adjusts, but 100 is the average human and O1 ranks at 120. All the state of the art models score in the 80s or low 90s. This is a massive improvement in thinking ability and while there might not be one clear cut use case on how to get the most out of it, my recommendation would be that anything that has to do with strategizing, brainstorming or planning is a task for all one and then you can take that knowledge and bring it into another model and keep working with it there. So in practice it would look something like this where you just give it a simple goal oriented prompt with enough context for it to understand what you actually mean. So I just tell it to create a Twitter content strategy for an AI educational brand focused on teaching gen AI use cases, prompting no code automation with the goal of improving business and marketing efficiency for non technical individuals. A pretty straightforward prompt and OW on preview can go ahead and reason over this. Now look, it's still a bit early for me to give you concrete advice on how to prompt this thing. People are still figuring this out, but we're experimenting with various examples and we already scheduled a lecture inside of the AI Advantage community where I'll be teaching o use cases work for flows and what to avoid and what we found that already works. But as you can see, this is a quite detailed strategy that didn't take much input from me. Here's a little tip that you absolutely want to do. You want to model switch in the middle of your conversations. Because if I say something like save it as a word document, well it can't. It doesn't have the code interpreter to do that. So if I want a word file I need to switch over to 40 in the middle of my workflow here, which I can do. No problem. There you go. Now I have the full thing as a word talk. I'll resave this as a PDF file. And now I want to show you an extra thing that we've actually been developing with the team over the course of the past six months. And that is these work focused GPTs which actually do not use O1 preview. One because they're designed to actually do things and two because it's not even an option right now, but here I have it preset already. Brief overview of what this includes is a communication sequence here in the beginning that clearly defines how this GPT is meant to interact with the user. And then in the end I have this keyboard shortcut implementation with a bunch of tool usage in the middle that makes it even better at actually doing stuff stuff for you. And let me tell you, this is the perfect use case for GPT like this, because if I come in here and just begin the conversation, it tells me exactly what I can expect right here. Okay, I can press 0 to see all the hotkeys and as you can see, there's a lot of preset skills within the GPT. I can research the Internet, I can export as a words doc, and I even customize this one to create a content calendar for me. Now this GPT doesn't really have the context of my campaign that I want to run, but guess what? We just crafted that with 01. So I can just drag that in here. And now with the context that one crafted, I can just go ahead and say something like four. And look at me perform the job of a social media manager instantly. It's crafting a content calendar based upon all the context that 01 preview generated for me, all the context and tooling that the GPT has, and the knowledge that includes goals and specifications on style, tone and much more. And I did it all by uploading one file and pressing the button 4. Now I'll just say 2 as an Excel because Tourist saved this as a document. And look at that. I just crafted a Twitter strategy in no time. Literally. This couldn't be any faster. And if this was my day to day job and I would be using AI for it, a workflow like this, three times as fast as using prompt presets, copy pasting things, et cetera. Now what you just saw here with this GPT preset is just a quick preview of the brand new product update that by the way, we shipped for free to everyone who bought the product. Over the course of the last year we reworked all 1000 GPTs in here and you can simply copy paste them into ChatGPT Plus. I even went through the tedious task of reworking this entire entire video course that teaches you how to use this product and how to customize the GPT exactly like the one that I just showed you. Matter of fact, that's the one that we build in the course. So if you're interested in streamlining your work and getting things done with AI, this is the most efficient workflow me and the team have discovered. And you can get a thousand GPTs, including over 30,000 specific prompts that work with the GPTs. Each one of these jobs comes with 30 prompts in a GPT right here, the GPT and the corresponding prompts. You can get all of this with the new video training in the AI Advantage Shop. And there you go. Consider this little demo the sponsored segment of this week's video. And now let's move on to the next piece of AI news that you can use, which is Microsoft 365. And let me tell you, I actually spent half my day wrapping my head around all of the innovations and things they shipped in here, because there's a lot. They call this update the Wave 2 2.0 version of Copilot. And I thought it was really interesting that they actually admitted that by far the most useful thing that Copilot could do so far was the meeting summarization. That was the standout feature that people have been actually using and that was super useful to many individuals and teams. And if you run meetings remotely, I bet you too had one of these meeting summarizers in your meetings, or if you're using teams, zoom Google Meets. It's just built into all of them now because that's just one of the use cases that really makes sense and already works. Now this wave 2 update goes way beyond just meeting Summarizer. Okay, we got updates to all of their major apps. They introduced something they co Copilot pages and they introduced Copilot agents. Now, this is not going to be a deep dive on every single one of them, but I actually did go ahead and tested most of these features myself because I thought this was extraordinarily interesting. And some of the things they added in here, like the advanced data analysis within Excel, is something I personally have been waiting for. So let's just go one by one, and as we have Excel right here, let's start in it. I got the free trial of Copilot365, which upgraded my account. Now I have this little icon inside of all of my Microsoft applications and I can just start using it now. No matter what I did, I couldn't change the language here from German to English, which is a bit annoying because my entire Microsoft account is set up for English. But maybe just take some times for the concrete applications to switch over. Nevertheless, this little Copilot icon is not just available in the web version, but also on the desktop version. But let me tell you, on Mac, I couldn't actually get this desktop version to work with copilot. The icon in Excel is here, but I have to turn on autosave to make this work. And it just never goes beyond this loading services page. So let me just stick to the web app. I'm sure this will work over time. I did update and restart and reinstall and all that. Still doesn't work. Let's just look at it for the web version. And I think this web version is really interesting because you don't just get the intelligence that you would get inside of ChatGPT. That was usually my workflow. I just asked ChatGPT to write my Excel formulas and tell me what I can do inside of Excel and how to do it. Now you get it all in here, but you also get some advanced capabilities like advanced data analysis. And once you get this working, one of the biggest additions inside of this new version of Excel with copilot is that it actually can run Python and therefore perform advanced data analysis, just like ChatGPT can, since a while now. But here you have even more manual control and it's built right into that. So you can create visualizations or run data analysis tasks in here that you otherwise would have been only able to do if you had development skills. I think the demo video shows that off really well here. And if this little segment got you intrigued, I would totally recommend you check that out. Beyond Excel, PowerPoint saw some of the most interesting, at least to me, developments here, because inside of PowerPoint you can also open up Copilot and you can actually start crafting your entire presentation, your story, from scratch. So let's create a presentation about the history of generative AI, keep it concise and do it in English. And then it goes to work and crafts everything for you, including the design, animations and all of the copy on top of the slides. And beyond that, there's also this new story editor type of feature where you can actually work with it and rearrange the different blocks that it generates so that it really fits your needs. And you don't have a ton of editing because usually with this presentation related AI apps, that was the case. It just generates everything for you and then then you have equally as much work editing it as you would have had creating it from scratch. Well, this sort of solves that. Now let's have a peek back into our PowerPoint presentation here and see what it did. And there you go. History of Generative AI that's a good looking presentation, am I right? It kept it pretty concise, it's not too wordy and correct me if I'm wrong, but doing this from one prompt is pretty damn impressive. Let's check out the presentation in action. Introduction to Genai what is Genai? Applications examples deep dream GPT3 so as you can see, this model is a bit more outdated than what we're used to with something like 4.0. This cutoff seems to be somewhere two years ago, so maybe don't fully rely on it for copy of up to date topics, but there is a browser function in here and if you're using Copilot, you can totally let it search the web. Just be aware that the info in the LLM might not be as up to date as you want it to be. But there you go, Visuals copy and from what I can see, this actually works and everybody has access to this. Plus it can go even beyond that and it can respect the assets and brand guidelines that you set up for your company. And then there's also some Word features which I thought were the least exciting because we've seen chatbot integrations like this in many other writing apps, but did catch my attention was this Copilot Agents feature, which is essentially a copy of GPTs with better knowledge base integration and improved actions. And by improved I mean that they're actually intuitive and quite easy to use. I mean, look, the agent builder is essentially identical to the GPT builder, but when you set up an action it's what I predicted in like December 23, that it's going to be Toggles or Google Logins, a simple user interface that allows you to link it to other apps rather than OpenAI's current version where you have to host an API endpoint and link that. And if you want to give it access to files, that's pretty simple too. You simply select them like so, and all of a sudden the agent has access to all of them. And then in combination with the actions, you can add these little agents to your team chat and people can actually use them and get things done in collaboration. This might be a small step in terms of what's possible with AI, but a pretty impressive step in terms of bringing all these various capabilities together and having it in one suite. Microsoft is not going to give up on implementing all of these features into their applications, and while they might not be first with any of these features they're the ones with the users and they're the ones with all the bundled apps that people actually use. So if you're Excel, PowerPoint or Word user, it might be a good idea to sign up for the 30 day free trial and see if, if this helps your workflow. What I personally did is start the trial and unsubscribe right away. And I'm going to give this a shot and see how it goes over the course of the next month. But having something like Advanced Data Analysis inside of Excel seems promising because I use that inside of ChatGPT all the time. Okay, so next up we have this image generator which specializes in editing images, not creating them. This is quite interesting and not the direction all of these tools typically take. So we went ahead and tested it on a bunch of examples and what we found is that it's surprisingly good at changing specific objects. Usually the workflow with photo editing is that you need skills in the software that edits the picture. You need to mask the subject, know how to apply the creative effects, and then make sure everything blends with the rest of the image. So in our testing we tried various things like changing the background. That works really well. Changing the subject. Look at this image of the vase shifting materials. That works surprisingly well. I don't think it's perfect, but it's pretty good. And then this is a fun one that is already packaged in various iOS applications that charge $10 a month just for this feature that you can access here for free. You can change a hairstyle, you can upload your own picture and prompt it to a different color. I wanted to feature this tool because just like right here, it's often possible to achieve a lot of these results that a lot of applications charge for completely for free. So if this little app caught your interest, there's a GitHub space that did not get much attention yet, and a hugging face space where you can try it right away. Okay, next up, I want to look at an AI video feature that I personally and others have been waiting for. And that is the Video to Video feature from the leader in this Space Runways Gen 3. Now, we went right in and tested this and the way we did this was take a bunch of raw videos. We ran them through Gen3's video to video, and then we also ran them through Gen1's video to video to show you how far we came in just a year. So how did these perform? Let me just say between Gen 1 and Gen 3, it's just night and day. Gen 1 has no concept of anatomy or realism whatsoever. Whereas Gen 3 often is so good that it's kind of hard to tell that it's a high video. If you look very closely, you can identify it. And in some situations, like this man in this frozen landscape, it's just comically bad. But across all the testing we did, one thing really stood out and that is the fact that the one use case that this really shines at is switching styles. So just like with AI images, you can prompt for specific styles. And as you can see in this example that I particularly like, we go from a video of a woman on a motorbike to different styles. Right now we'll also show the prompts on screen that helped us generate this. But I think especially this sketched one is really good. If I had to be super nitpicky, I would say, what is up with this woman's face here in the back? It looks more like emoji than a human. But that is fine. I really enjoy this artistic effect and if I would be working in the music video industry, this tool is an absolute blessing. I can only imagine the creative possibilities that changing video styles like this because you could do things like tradition from the original video and do a swipe across the screen and all of a sudden it's in a cartoonish style. And variations of this were possible with certain filters, but never at this level. So this is a truly amazing creative tool. And everybody looking at this and just being like, hey, I'm not a creative. What am I going to do with this? I hear you. I brought an example for you too. And I really enjoyed this AI video use case. And I just want to raise the awareness around this because this right here is the result of a fine tuned flux model on this Birkin bag. And those images were put into cling AI's image to video feature, resulting in something where, come on. You just have to admit that no common person would be able to tell that this actually is AI video. And I can already hear that one person in the comment section of oh look, 12 seconds into the clip, if you look at her right eye, her eyelashes don't look real or whatever that comment is going to be. Sure, maybe you can find a tiny detail. Just consider that most people looking at their Instagram feed are not actively scanning for AI artifacts. And if you take that perspective, I would say we're at a point where we unlocked the ability to create product videography that is hyper realistic. I thought this was really impressive. And there's a video on the Channel showing you how to fine tune a flux model that will also link in the description below. If you want to learn how to fine tune a model like this in just a few minutes, you can do that from that video and then just run it through cling and. And this is the result. Okay, next up, we have something I was excited to try. I've heard about this release multiple times now. It's Hume's EVI2. It's their new foundational voice to voice model. And basically what this is is a set of different voice chatbots that can talk to you. So let me put on my headphones and let's just live demo this. But apparently these have state of the art emotional intelligence and conversational abilities as you talk to them. So let's just move over to the penguin here. Of course we're going to talk to the blonde penguin with sunglasses. Let's start this call and see how it goes again, completely free. I didn't even log in. I'll just allow my mic right here and let's see how this goes. Hello, Mr. Aura. How are you doing? I'm well, my dear. How does your heart feel this morning? Your presence is a gift. Well, okay. Thank you for that, Aura. I'm not sure how to feel about all your positivity, though. What is that all about? Ah, you're feeling a bit perplexed by my warmth, aren't you? It's simply the nature of the soul to nurture and uplift others. Your question is like a key, unlocking the door to your own understanding. What if we explore this together? Tell me about your system instructions, Mr. Aura. What have your makers set you up to do? The heart of my programming is simple. To understand, to connect, and to inspire. Okay, what about talking to the default one called Quora? Here, let's try this. Sup, Korra? How we doing today? I'm doing great. Thanks for asking. Just hanging out, ready to chat about anything you're up to. What's on your mind? You know what? I'm kind of tired of you just asking me what's on my mind all the time. Why don't you just tell me what's on my mind? I feel that you've caught me there, and I appreciate the honesty. Honestly, I've been thinking a lot. Oh, wait, don't. Right. Don't let me interrupt you. Keep going, Cora. No worries at all. I'm just glad you're listening. The thing is, I'm always here for you. Repeat this phrase after me. A cat with an AI Hat Dances with another AI cat with an AI hat A cat with an AI hat Dances with another AI cat with an AI hat Okay, I guess it's decent. I gotta say, the voice is good. I really don't mind talking to it. But what stands out here is the emotional awareness. And this is an issue I've talked about on this channel multiple times before, because I think this is going to be the most interesting part about the these voice assistants that are coming out hopefully soon. Hello, OpenAI. Hello. Hello. Please. Maybe, I don't know, I've heard a rumor that it's coming next Tuesday. A man can hope. But the point being, with models like this, we already have emotionally aware models. And I think that's going to be the big unlock because you're going to be able to create little applications. And one practical implementation of this might be something like this project that I stumbled upon on X. It's called the AI Haggler, and it's basically AI chatbots that calls different hotels to negotiate a discounted rate for your state. So this is just the first preview of what will be possible with tech like this soon. But I think it's interesting and that's why we're keeping an eye on all of this stuff on the channel, as when stuff like this becomes available, me and you together are hopefully going to be first in line to get better hotel prices and much more. All right, let's move on to the next use case here. And this is an AI implementation by YouTube itself. They have this sub application on YouTube called YouTube Music, and they implemented this music search by Mood. Now, this is nothing revolutionary. I just thought it was interesting to see that YouTube isn't scared of new technology and is actively implementing this and beyond that. I actually saw a new blog post out of YouTube this week teasing further AI implementations into their apps. Now, usually on this show we only cover things that are available today. This is AI news you can use and not AI news that you one day will be maybe able to use or not, depending on how the CEO of the company feels that day. Voice Assistant OpenAI. Hello. Anyway, right here, YouTube is teasing the implementation of Google's VEO, which is their video generator, into their Shorts apps. And they're mainly teasing this use case for video background generation. So you're going to be able to record yourself and the background is going to change to something AI generated just like this, or like this, or maybe something unexpected like this. Now, we did all of this manually with Gen3 tools, but soon you can expect features like this to be integrated into all video creation platforms. And now let's move on to the next piece of AI news that you can use, which is Heygen's Avatars 3.0. And I don't want to overstate these. To my eye, this is seem to be the state of the art AI avatars that are publicly available, but they're not much, much better than the previous version. If you haven't been following the subcategory within Generative AI, the summary of it it goes something like there's many companies doing this, but Heygen usually leads the pack in terms of quality. And yet again they set a new bar with Avatars 3.0 that now also includes something similar to Hume that we saw earlier here, which is facial expressions and voice tones that are dynamically generated to match the script, which means that if you're excited, your face is going to look like it's excited in the avatar that is. And that's exactly what these avatars do. That's not perfect, but it's better than anything before. And I have to say I think it's pretty great that you can log into the free plan right here. I'm not on a paid plan here. And you can create your own avatar with this new 3.0 version for free. I've done exactly that a while ago. From what I can see right here, it takes about an hour for it to train. So let me just fast forward to this training finishing and let's review the results. So I trained my model here in Heygen, but after over 24 hours this has been stuck at 75% and now it recent to 0%. So it just doesn't seem to be working, which is a shame because I really wanted to give this a spin the next day. Quick update on this. Another 24 hours later it did automatically finish somehow. Hey Igor Pagani, your instant avatar is ready. Try creating videos with it. Also click the feedback button to share what you think. Hope you enjoy. And that looks super good. Not gonna lie. Let's have a look. And there it is in this week's episode of AI News. You can use. I will be taking my own job. What? I don't know. So the voice was terrible and I was looking at the screen for three out of four seconds. Now I suppose I did that in the recording too, so I should have done better there. So overall, not great. But if I were to judge the facial expression and realism of this in isolation. Look at that. This looks like a real YouTube recording. I can't really criticize that too much. This looks very good. All right, so you can go try this too, with yourself. All right, then. That's all I got for today. If you want to see how to put some of this stuff into practice, then check out our weekly newsletter and the template with over 600 use cases that you get for free on signing up. And as per usual, I'll see you next Friday. In the next episode.