# 9 INSANE Use Cases With OpenAI o3

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=Sq0VCbGdRJU
- **Дата:** 22.04.2025
- **Длительность:** 17:09
- **Просмотры:** 21,499

## Описание

Join my AI Academy - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Links From Todays Video:


Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

(For Business Enquiries)  contact@theaigrid.com

Music Used

LEMMiNO - Cipher
https://www.youtube.com/watch?v=b0q5PR1xpA0
CC BY-SA 4.0
LEMMiNO - Encounters
https://www.youtube.com/watch?v=xdwWCl_5x2s

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

## Содержание

### [0:00](https://www.youtube.com/watch?v=Sq0VCbGdRJU) Segment 1 (00:00 - 05:00)

Now, Open Eyes's new model is here, and I think this is absolutely game-changing. Now, I know I usually say that all the time, but this one is completely different, and I want to make this video showing you guys the absolute best use cases that I've personally found that I'm incorporating into my workflow that I'll use on a day-to-day basis. So, without wasting any more time, let's actually dive into the best use cases for OpenAI's 03 model. Now, before I start this video, I want to make something clear because this only became clear to me as I was really testing the limits of this model. One of the limitations of 03 currently is that even with a chat GBT plus team or enterprise account, you only have access to 50 messages a week guys. That is 50 messages every single week with oree. That means you get access to just around seven messages per day. So whilst using this video, of course, try things out, but don't go overboard because most people don't realize that there are really low limitations to how far you can push the model even with a plus account. So, just wanted to make that clear. And if you do start to get close to that limit, it will actually notify you about the different use cases. So, now the reason that 03 is actually so different and the reason that the use cases in this video are a little bit different is because 03 is a model that essentially is an AI agent. This isn't just an LLM model anymore. What they've done is they've designed 03 from the ground up to be completely agentic. This is why a lot of people are saying the model is AGI and that's why a lot of people when they use the model now they having much more success because it does so much more. And I wanted to set that preface for you guys so when you're looking at these use cases you can understand why it is the case that the model is able to do certain things. Remember guys the future is agentic and what we're leaning towards is a model that can pretty much do anything. And in this video, you're going to see exactly why 03 is just, you know, more agentic than any model out there and why you don't have that many messages you can use. So, one of the most popular ones right now is advanced image reasoning. We all know that they released a set feature called thinking with images. And essentially what that means is that you can basically reason with your images. So thinking is of course something that is allowing you to have greater intelligence when it comes to having difficult questions being answered. And essentially what we have here is we have advanced image reasoning. So what that means is that it can look at an image and reason throughout things that it sees. Now it doesn't just mean that it looks at an image and it's able to just immediately identify things. What it actually means is that it's able to zoom into different things and it's able to look at key details. So let's say you want to ask a model a question about an image. What it actually does is it doesn't just take that image, analyze it as one. It dissects that image and reasons about that image with a ton more information. So what the model will do is the model will essentially think about the things that it's seeing. It may look at the scenery. way the sun, you know, overlooks the shadows. It may look at the housing in it. It essentially also dissects the image as well. Now, what's crazy about this is that it actually got it right. actually just typed in Las Vegas and I saw someone take a picture of Las Vegas and it actually got this right, which is pretty crazy. And I'm gonna actually show you guys the reasoning because right now I don't know why it didn't show me the reasoning here. It usually does when it thinks for longer. It usually will give you a really nice reasoning trace. For example, this is something that I found on Twitter and basically the reasoning trace is showing you that it analyze the image in several different ways. So it will zoom into you know different parts of the image and then you can see it says these mountains don't seem very tall so I'm reconsidering my guess yada yada and so what the model is able to do is it's able to really pick out key details unlike other image detection software you know other models with vision and that allows it to be much smarter when it is guessing or trying to reason about different places and locations. So, one of the things people have been doing is just inputting random images and then trying to basically see where it is. And this is something that some people are essentially calling location AGI because you can put in an image with almost no real details about where exactly it is and it will just simply analyze the image. So finally now you can see it's you know showing exactly what it's doing. So if you expand this, you can see that it is analyzing the image in a specific way. And you can see that it is, you know, giving us the thoughts on how it's analyzing that image. And the reason I also like this is that if you're really trying to analyze an image for something, it might not be location focused, which we'll go on into a moment, but you can see what it's looking at and what it's paying attention to. So right here, like I said, it's going to zoom into this part of the image. And you can see that it is you know reasoning about where things could be based on you know how it's extracting information from that image and you know cross referencing that with its internal world model. And then right here we can see that it is also searching the web because um you can see right here like if we go on to this you can see that it was like okay based on this I'll need to do a quick confirmation. So to confirm I'll quickly search for drone images of California city. And then you can see right here it

### [5:00](https://www.youtube.com/watch?v=Sq0VCbGdRJU&t=300s) Segment 2 (05:00 - 10:00)

gives all of this information about this stuff. So like I said remember guys this is an agentic model. It's not just a model that you can ask something and it's going to reason with itself. Like I said, it takes the image, it uses tools to, you know, change that. Then it reasons with itself and it thinks, you know what, let me go and have a Google search and then it gives back the information in a table. This is exactly what a human would do and this is why this model is so powerful and this is what I'm trying to get across to you guys. This is probably the most underrated model at the moment. Now, of course, with advanced image reasoning, some people are using this to solve puzzles. Although one of the funny things that people are doing is using it to solve where's Waldo puzzles. I know it's not really a business use case, but if there's anything difficult about an image, maybe you're trying to solve a riddle or just anything really difficult that you can take with a picture, this is going to be something that you could, you know, have the model do. So, like I said, it's going to be able to zoom in, look around the image, and it's going to be able to give you a nice answer. Now, 03 and the OpenAI models are essentially quite confusing because we got search here. Then of course we've got deep research here which you can use with you know uh you know your standard model like GPT40 and of course you can use you know deep research with 03 and 03 will essentially do deep research by itself like it's basically instructed to do that. So if you want to research anything this is basically something that will give you deep research but it's going to be a lot quicker. Now, of course, you might want to use deep research if it's cheaper because 03 can do a lot more than deep research, but if you wanted to research a topic quickly, like you needed a research report very quickly, this would be the way to go. Now, the next one that's really, really cool is that you still have access to Python/code interpreter. So, what it's able to do is it's able to, you know, put stuff on images in the sense that it's able to reason about things. Basically, you can solve mazes. Okay? And I'm sure there's more use cases, but this is one that I found where, you know, you can draw on images, which is really cool. So, you can see this person, and I'm going to show you guys a real demo of this. This person found that, you know, they used 03 to find the path through a 200x 200 maze in one try. And of course, they had to double check the solution, but it actually works. Now, I wanted to test this out myself, and so I did. I actually just grabbed a random maze off of Google, and I put solve this maze for me. You can see that the exits are here. The entrance is down here. And then you can see it thought for a minute and 43 seconds. And once again, you can see how it reasons with the image. It says looking at the borders. The left border seems to have a thick edge. There might be a gap here. You can see it says this is thick. It looks down here. And then eventually you can see it reasons. It reasons. And after all of that reasoning, it manages to find the entrance and exit. So it gives me the final image here. You can see that you know the entrance is here. It managed to go all the way through, all the way around, just sticking to the borders basically. Basically like a kind of sticking to the left strategy and then coming out at the end right here. And this only took it a minute to do, which is pretty cool. I don't know what use case you would actually have for this, but um I'm pretty sure that there's going to be some really creative use cases. So, this is something that, you know, I do think it's more of a fun one, but nonetheless, it's still a use case that I do want to have here because it seems really useful. Now another use case that is probably a little bit more on the useful side is business analysis. So basically what I did is I took some synthetic data for a Shopify store and I basically said look analyze the data for the last 3 years and then give me a historical projection. So let me show you guys what 03 was able to whip up. So you can see right here I said I have a friend who's been running a Shopify store selling vintage clothes online for the last 3 years. Can you take a look at the data, analyze it and let me know future projections based on any trends you see or any tips and insights and create future projections chart based on your recommendations. So once again it thinks it imports it. It then you know has all of the data and then it says it looks like there's a compound annual growth rate at around 1. 3% which is positive but not huge. Then it's looking at, you know, seasonality. I should focus on monthly trends. And then it's like, you know, to really dive into the data. I should focus on data that might be around March or September. And basically, it just analyzes all of this stuff. And then right here, we do get a nice graph. So, you can see it's giving me projections based on the future that do tend to line up with the graph right here. You can see this graph was slowly going up, which is really nice. And then it also gives me some, you know, really nice advice right here. So this is really nice because you can see the historical data, the forecast data and then when the forecast starts. So this was something that was super useful. So you can see right here it says you know what we see it's slow but positive growth. Business is stable but not compounding fast. You can see it says right here set a concrete target 10% yearon year and track that you know that growth. You hear sharp seasonal spikes revenue peaks in March, June, September. You have dips in January, February and midsummer. You know shoppers buy vintage ahead of the festival season. summer holidays, it's back to yada yada. It says concentrate ad spend on drops six weeks before those peaks. Build that for, you know, to raise average order value. And all of this stuff is basically allowing you to just run your business in a more effective way. I mean, consultants charge a lot for this, but I mean, giving your, you know, all of your data to a model and saying, look, generate a chart being, you know, just just

### [10:00](https://www.youtube.com/watch?v=Sq0VCbGdRJU&t=600s) Segment 3 (10:00 - 15:00)

let me look at that stuff. It's going to be really, really interesting. Now, this is something where you can see once again, it's giving you a 24-month projection. It gives you, you know, things that you need to do now to lift the potential. It's really, really interesting. And you can see it also gives you some tactical tips. And then a really nice thing as well that, you know, I wanted to talk about is the fact that with 03, what you actually get access to is even faster deep research. So I'm actually going to use that in conjunction with this. So essentially, take a look at this. Basically, OpenAI showed us in the benchmarks that 03 with Python and browsing actually gets basically the same performance as deep research. There's only one caveat to that though. Remember how at the start of the video I basically said that 03 is essentially a lot more expensive than deep research. So if I go back to the tab you see that with deep research you can see I've got 114 available until May the 2nd. But with of course the 03 model I only have 50 messages per week. So I have to ration those a little bit more. Now deep research is going to be a little bit longer and a little bit slower. But what you can do is you can now that you have that information, I can say, "Okay, look at this. Look at online trends and then come back with a research report on actionable things I could do for my store or anything that's upcoming that I could take advantage of. " So now I'm going to give it that. And now it's going to go ahead and think. Honestly, hope I don't run out of, you know, 03 uses during this video, but it's all for demonstration purposes. So now it's going to think, it's going to reason, and it's going to go ahead and give us an answer. And so now that we've used this quick research, I didn't use the deep research, but you can see right here it went around search the web and then you can see it managed to give me some decent information. It was going on all of these sources. It could it was like, you know, now it's summarized to summarize these trends and steps. And when we also dive into the thoughts, we can really begin to see what the model, you know, is focusing on, how it's directing it. And I would always advise you guys to do this because one thing that I've realized is that when I prompt a model, oftent times my prompt lens is false. What I mean by that is that I'm thinking in a certain direction. And unless the model is thinking in that exact direction via my prompt, if that doesn't occur, then I'm not going to get a good response. So right here, you can see that this, you know, it does have the right lens. It's thinking in the right way. It says, I'll search online for insights that are directly tied to these topics. topics such as vintage clothing, yada yada. Location isn't necessary for this task since all trends are global. So, I avoid focusing on fashion based details and concentrate on the broader e-commerce trends. And then we get down to here. We've got all of this data. It says run weekly live Tik Tok shop streams on Tik Tok shop or Instagram live, you know, shopping adds 36%. This is actually true. I know about this is true. Crosslist your catalog. It gives you all of this information. So, this is stuff that is real and I think really nice. You can see right here it says quick scorecard for imple implementation. It gives you the cost, the effort, uh the expected boost. And I mean imagine you're someone that runs a store. You could easily, you know, break this down with your team and it gives you the next steps. Okay, you know these these high impact lowcost tactics. I mean stuff like this is absolutely insane. Like having this within seconds that the model was able to do it. I mean the future is honestly going to you know belong to agentic individuals, individuals with high agency. So, this is pretty crazy that you could do this in just seconds. And this kind of research is going to be a little bit better than deep research. So, it's what I would use. Now, something strange as well that 03 is actually good at that I'm not sure why it's good at this, but it actually is creative writing. So, there's this benchmark called creative writing V3. This is the emotional intelligence benchmark for LLMs. And apparently 03's, you know, score is pretty much number one. So, this is really surprising to me because it's basically set up to be an agent using tools, but somehow it manages to achieve number one on the creative writing v3 score. And I would have thought that GPT4. 5 would have been the model, but it seems like for some reason we've got two reasoning models that are at the number one spot when it comes to creative writing. Maybe the way the questions are phrased, maybe that's probably why. But I'm guessing that this is something that's remarkably interesting. Now, I will say the obvious here. Don't use 03 for creative writing because it is incredibly expensive. Maybe just use GBT40 or DeepSeek R1 because those are way cheaper in terms of what you're going to be able to get. But I just wanted to include this because I know some people still do creative writing and this is going to be a use case that some people may want to have. So for those of you guys that do have that use case, understand 03 is number one. However, there are much more cost-ffective options and I don't think this is what they intended the model to be used for. Now, there are some limitations of 03 and the reason I'm including them in the video is because I think it's rather important to actually point out where the model has limitations. Often time we get caught up with just how good things are, we overlook things that are quite bad and sometimes can ruin our pretty good use cases. So sometimes even though thinking with images is really good, counting things isn't something that it works well with. So being able to look at you

### [15:00](https://www.youtube.com/watch?v=Sq0VCbGdRJU&t=900s) Segment 4 (15:00 - 17:00)

know intersections in lines or be able to count things. This is just not something that is a strength of current vision models in LLM. So right here this clearly has 1 2 3 4 5 six fingers. Okay. And then you can see right here it says the emoji shows all five digits. Four fingers plus a thumb. So five in total. And this is something that happens you know quite a bit. And you know it basically just over reasons. Like even here it says I can see five upright fingers plus a thumb. That would make six digits. But depending on the emoji style it might seem like there are six digits but generally it's five. And it basically just gaslights itself into thinking that there's five because there's usually five. But in this example there are six. So the point is here is that this is overreasoning and it's hallucination. Okay? Like sometimes vision models don't really see well but sometimes they also hallucinate. And this is something that you need to know because in the research paper that OpenAI dropped with this, they essentially talk about the fact that 03 and 04 mini, these reasoning models that are much smarter, unfortunately, they do hallucinate a lot more. We can see on this person QA benchmark where they tried to elicit, you know, hallucinations from LLMs, 04 mini hallucinated 48% of the time and 03 hallucinated 33% of the time compared to 01 which was 16% of the time. And that's relatively high when you factor in a lot of the tasks that these models are going to be used for. So if you're using this model, please make sure you double check everything because if you don't double check everything, you might be in a position that is quite unfavorable considering that, you know, it's doing a lot of work. You just need to, you know, cross the tees, dot the eyes, make sure that the numbers are all perfect. I know it does seem a little bit annoying that you have to fact check something that's supposed to be saving you time. But if you're in an industry that really cannot have any mistakes, this is something that is a non-negotiable because these models are known to hallucinate. Known to hallucinate so much that they're able to basically lie when confronted with certain things. 03 basically lied about running code. And I mean, there was just an entire video that I done on that. Hopefully you guys enjoyed these use cases. I'd love to know what you're using them for on a day-to-day basis. And I'll see you guys in the next

---
*Источник: https://ekstraktznaniy.ru/video/12986*