what are the 01 models actually used for? What are the use cases? 90% of use cases can be handled by any model that you can fit on your computer. Then another 6 to 8% of use cases can be handled by GPT-4 or other Frontier models. And there's really only about 1 to 2% of all use cases that really require absolutely Cutting Edge intelligence. And that is what I'm going to show you in this video. Now, this is going to be a little bit of a different video, and I'm going to give you a peek behind the curtain for a second. I actually started this video as just a simple reaction to the demo video that OpenAI just put out about 01 use cases. I was super excited to review it. I wanted to talk about it with you. But then something happened. I started going through each of the use cases in the video, testing them using GPT-4 rather than the quote-unquote thinking model. And so let's find out what happened. All

Right, so I'm going to skip the basics because basically, they just talk about where the 01 model falls on the curve of intelligence and their releases, but we already know about this stuff. Here's a quick glimpse at it. Let's keep going. All right, here they're going to show the difference between the GPT-40 model and the 01 model. Where the GPT-40 model just gives you the first thing it thinks of, you ask it a question via a prompt, and it just replies with whatever it predicts as the next token. Whereas the 01 model takes a moment, it thinks through it, it outputs, it reflects on that output, it uses Chain of Thought, and then finally, when it comes up with that final answer that it believes is right, then it shows it to you. So, let's see how she describes it. And the 01 series works really differently from GPT-40. The most obvious difference you can see here is response speed, and the 01 series really marks this change from predictive responses to logical problem-solving.

given their ability to think, known as reasoning, and through training, the 01 series learns to refine their thinking process. They can try different strategies, and through that process, they can even recognize their own mistakes. You can see some of that in the summary of The Chain of Thought, which Joe will demo in just a little bit. Okay, again, this is all introductory stuff. Let's skip ahead again because I want to get to the use cases. So next, Joe from the solutions engineering team talks a little bit about what is in the 01 model and what is not. So a few things that are not: it doesn't have any tool support, it has no access to the internet or browsing, it does not have GPTs, so custom GPTs, and it has shorter context windows. Now, it seems obvious to me that all of these things are going to come soon. Now, can you imagine 01 with tools or browsing? I mean, it would just be so powerful at that point. I cannot wait to see that. All right, so here is the universe of what OpenAI sees as the

areas that OpenAI's models can actually add value. Val content generation help. Des strategy coding research data analysis translation and task automation. Now he's about to touch on where specifically the additional reasoning capabilities in the 01 models come in handy. So let's watch today. Teams use Chat GPT for several valuable use cases, from content generation to data analysis. During this research preview, we've learned from our customers that in the 01 series, it adds the most value in three areas: strategy, research, and coding. Let's talk about this, what this looks like, starting with strategy. OpenAI 01 preview can break down complex problems into steps, consider different solutions, and walk users through its process. This makes a strong ideation partner on strategy development, in business operations, engineering, and more. So first with strategy, if you're thinking through hard problems, complex problems, they are saying that the 01 model is especially good at helping you think through those complex problems, strategy being one of them. Strategy being more amorphous, not really something where you ask a question and it has a definitive answer.

to it, but rather something where it actually has to use some logic and reasoning to think through the possible outcomes of the strategy. So, here are some examples in business operations and engineering, such as new product introduction, pricing strategy, marketing campaigns, some from operations, organizational structure, supply chain management. These are extremely complicated use cases. Engineering, we have cloud migration, DevOps strategies, etc. So, one thing that I've said already a bunch about the AI models and really just advanced intelligence in general is that for 95% of use cases, you really don't need the AI models or that level of intelligence. However, for the use cases where you do need it, those are incredibly valuable. So, I'm not saying that the AI models and that level of intelligence is not valuable; it's just valuable on a much more narrow set of use cases, but those use cases are extremely valuable. Let's keep watching; we'll walk through a market expansion scenario in a live demo. Secondly, AI mini is especially helpful with coding tasks, from writing code to other adjacent tasks for engineering.

teams, yeah, and again coding one of those really hard problems even for large language models. Large language models code really well, but it could be so much better, especially if it's able to reflect on its own code, fix bugs before it actually gives it to you. One of the things that I do with coding and AI all the time is I ask for some kind of method, I test it, I look for the error, paste it back in, and then iterate on that feedback cycle again and again until I get something that works. O1 kind of shortcuts that entire cycle by doing a lot of that testing itself, or at least reflecting. It's not actually executing the code yet, but it is looking at the code as it's running inference and generating the code. So, some use cases that it handles: new projects from scratch, turning scripts into full-stack applications, SQL generation, code reviews, creating mock data, a whole bunch. Now, enough of this, let's get to the real-world use case. So, in our first example, I want to start with a strategy example in

this scenario. I want you to imagine that I've been tasked with coming up with an office expansion plan for my company to figure out where we should build our new office and in what market. In order to tackle something like this, it obviously requires a lot of real-world knowledge, which is why I'm going to be using ChatGPT-1 Preview as my model of choice for this prompt. Now, in order to get to my answer, I've given ChatGPT-1 Preview a little bit of context. You can see here the type of business that I work for, where do we already have some offices, where what seems to be one of the favorites so far, and some direction on some of the things that I want to make sure that it addresses. But I'm also leaving it fairly open-ended in case there are things that I might have missed. And to be honest, a question or a task like this has so many things to consider that ChatGPT-1 Preview becomes a wonderful partner to help you consider different aspects of

that challenge all right. So, I want to pause for a second. One of the coolest things that I personally do with any large language model is that kind of partnership process and coming up with the best response. I asked for something, maybe I didn't phrase the prompt properly or asked the question with all the context necessary. I see that in the response from the model, and then I iterate again and again until I get what I need. Now, what he's saying is that with the 01 model, you can also do that. But in my mind, because of the latency of the 01 model, that partner process seems to have a little bit more friction to it. So, let's keep watching. So, when I submit this to ChatGPT, one of the first things that you'll notice is that it kicks off its thinking and reasoning process. And if I expand this, you'll see each of the steps that it's taking in its chain of thought to start to come up with a plan for me to use for my office expansion and seal.

Notice gives me a very robust set of details around the market potential of Paris, but it also calls out alternate cities, places like Berlin. Maybe we'll come back to that in just a moment, and some of the other things that I specifically asked for, things like market entry strategy, how to build a brand cache as a thought leader, very helpful stuff, even calling out specific customers in the market that I can go to. But one of the things I really love about this that I mentioned before is that it gives me additional areas to consider, like acquiring talent in that market or building partnerships and alliances, even getting to things like financial planning and risk analysis. This becomes easily an area where I can think that there's many things I might not have considered and use O1 Prie to help me dig in further. So let's actually take this and do something that I've seen a lot of customers use it for, which is actually going a little bit deeper. So let's dig into BWI now. I'm thinking

immediately. How would the 40 model have handled this? Would it have been just as good or not as good? I know that it has access to tools. I know that it has access to the internet. So, in my mind, this is a pretty straightforward problem to give a model, and I think it would have done just as well. In fact, let's test it out. All right, so I grabbed the same exact prompt. Let's give it to GPT 40. Okay, advantages of Paris talent pool, AI ecosystem growth, government initiatives, corporate landscape, other considerations, Berlin, Amsterdam, Barcelona, B market entry strategy for Paris. I mean, this seems like just as good of a response. Yeah, so looking at the output from 01 preview and the output that I just got, it looks just as good. So, I'm pretty sure that the 01 model was not really necessary for this prompt. But let's keep watching. So, let's dig into Berlin as an option, and just like before, it will kick off its own thinking and reasoning process and go through all sides of that.

potential, uh, challenge. All right, so let's do the same thing, dig into Berlin as a potential option. Okay, so I'm going to let that run. Let's keep watching. I'll go into my next example where I want to switch gears into something, uh, that where we specifically optimize the 01 series for, and that's on coding. All right, so I'm going to pause there. Let's go back and let's read this. Look at this. So here's Berlin, a bunch of information. Yeah, I just don't think this use case really needed the 01 model, but I bet coding will. Let's watch that now. You notice in this, I've switched over to the 01 Mini model, and we've optimized the 01 Mini to handle challenges of round coding, uh, extremely fast and give a robust amount of detail. I really like this as a partner in coming up with addressing coding CH challenges and adjacent tasks because it gives me a really detailed step-by-step walkthrough so I understand each of the pieces along the way, um, and so in this.

scenario. I want you to imagine that I've been tasked with coming up with a new web application from scratch, um, with a couple of parameters: a Node.js backend and a React front end. Um, and I may be familiar with some of the concepts, but I haven't done this in a while, and I have a quick turnaround, so I'm going to use ChatGPT to help me come up with the plan for this. And so when I submit this one, one of the first things that you'll probably notice is that it still does that same thinking, but the speed at which it performs its reasoning and consideration, um, is much, much faster. And just for the heck of it, let's give that same exact prompt to GPT-4. Let's see how it does. So I'm just going to hit play as we check out the rest of the video. This is that still does that same thinking, but the speed at which it performs its reasoning and consideration, um, is much, much faster. Thought for a couple of seconds, and that doesn't mean, uh, it

detracts from the amount of detail. Quite on the contrary, we actually have a very robust set of information or walkthrough on each of the phases to actually bring this project to life. Project overview to actually getting the front end set up, uh, and running the application. And again, going back to this, I've used this in my own personal usage. I've watched customers use this. Um, this has been uh, an amazing helper in trying to understand each of the steps that I'm going through, all the packages that I need to install, how do I make sure that I've each of the file structures all right. Very impressive. Now let's see how chat GPT 40 did. Okay, so same thing, project structure. Here's how to initialize, installing the necessary packages. Here's some code, the app.js code. And obviously, I'm not testing it. I don't know how it's actually doing, but seems to be just as good in my opinion. Let's keep watching, correctly set up. I'm literally so in many cases copying and pasting some of this into my, into my

own environment, but I think the most important thing here at this phase is that I'm getting a really good understanding of what is happening at each step. And as I keep scrolling, you can see it's quite a robust amount of information. Um, and so I think the only other thing I'd want to highlight here is that, again, if you start to dig into other areas like, uh, let's, okay, let's pause, let's go back, and yeah, I mean, it also output a robust amount of information. Doesn't seem to be quite as much, but everything seems to be here as far as I can tell. Connect this to an Azure database. It's going to be a really great partner in even taking that new challenge and adding it and syncing it into the current, uh, the initial prompt that we had. Okay, so, so I'm going to pause, let's do it again, let's connect this to an Azure database. Now, the 01 Mini model is actually really fast. It seems just as fast as this. Obviously, I'm just eyeballing it, but it does. It just

It seems nearly as fast as the GPT-40 model, or maybe even faster. It probably just depends on the time of day. So here it is, giving me everything I need, telling me step by step how to do it. Okay, so that's the coding bit, and so far, 01 seems good, but maybe not absolutely necessary. Let's see what he has to say about the research use cases I mentioned. Uh, I'm not an expert in genetics, so I won't be using that example for today. I don't have my PhD, but what I like to believe is that I'm a very attentive dog dad. Um, and part of being an attentive dog dad is that I've been trying to research and understand the best way to give him a long-lasting life, and that's usually through creating an optimal food. And so, not being an expert in research, I can still use ChatGPT-01 preview to get insights on how I can learn more, more about ways to get him an optimal food, um, and also consider areas that I didn't.

didn't think about necessarily before, but I think one of the key things I'd love that I've found that researchers have used this for, that I'd love you to walk away with, is that it's really helping you automate a lot of the more long-form or laborious tasks that are in the world of research. Not necessarily coming up with novel AI innovations or new inventions, but it'll help you work through that and give you areas to explore new concepts. All right, so when I started this video, I definitely didn't plan on recreating all of the use cases that they show in this video in GPT-4. But once I started, it became pretty interesting, and let's find out, let's see if the 01 model is really necessary for this research use case. So I got the prompt once again, and let's see what it does. Let's keep watching while that's outputting. And so I know some things here. I've given it some prompts of how I want the goals that I

want I know a couple of things like Three fatty acids are important but I am open to other considerations and again helping me through some of those more monotonous tasks of creating research plans and who knows if I come up with something great maybe it'll even give me Insight on how I actually can make this into a product and bring it to other people and so when I submit this it much like in the strategy example take in a wealth of real world information combine it with our capabilities around science to come up with a very thoughtful plan to then give me a step-by-step guide on how I can create that optimal dog food now again going back to before I knew some things like omega-3 fatty acids were helpful but this was immediately open my eyes to things like the balance the ratio for omega-6 fatty acids um I also started to think of okay let's pause for a second let's look what chat PT did look at this it's basically the same output yeah so a lot of these are

exactly what was found in the 01 model. Gather feedback. Here we go. So yeah, product line expansion, everything's here. So let's see what the 01 model did, like using taurine, okay, prebiotics, and probiotics. Interesting. I take those, maybe my dog should too. So in the area of research, where I can go and explore those, but I also like that it pulled open each of the different phases, is potentially bringing this to a real product, including things like testing and validation, regulatory compliance. All right, maybe coming up with my own dog food line is not in the cards, but at least I have a path to go get a better dog food for my good pup. All right, so once again, I actually don't think 01 was really necessary. I think GPT-4 solved it just fine. For those of you that are familiar with ChatGPT or just LLMs in general, they've typically struggled with mathematics in the past, things like counting the number of R's in the

word strawberry has been very challenging. Um, not the case anymore with the R1 series. Now, in this example, um, I want to show you an example of how we're improving upon that and how we've actually optimized again 01 mini for T tasks that are related to mathematics. Um, and I'm going to be using a covered call option. Now, I just one quick disclaimer. Um, we don't intend for chat gbt to be a replacement for a financial expert. Um, we also don't intend for this to be your guidance in the stock market. Not investment advice, basically. But hopefully, it can... This can be illustrative of how we would perform some of these calculations. And the thing that I want you to make sure that you take away from this is you don't have to be a stock market expert to understand this. But um, knowing that you can still use chat GPT for any style of prompt, no matter what your workflow is, even if it includes mathematics, and still uh, feel like it's going to be very accurate in

its response, and so for this one, I'm looking for a way to maximize my profit. And again, because we're using the 01 Mini model optimized for math, it's going to reason through those different challenges quite quickly. Okay, so yes, LLMs have traditionally struggled with math, and I have also seen that the 01 models are really good at math, but let's actually see what happens here. And I must say, watching the 01 Mini output, at least on his version, so we don't know if he has kind of higher inference speed access or what, but it is lightning fast, maybe even faster than GPT-40. So let's see what it says. Maximum profit $130,000. If the stock rises to $17, you still make $130,000, but you forgo an additional $70,000. All right, so let's give that EX ex prompt to GPT-40. Okay, so GPT-40, and let's see what it says. So it's definitely slower. Your maximum profit is $130,000, and maximum profit

$130,000. You lose the opportunity to sell at $1,170 as you are forced to sell at $60. The $3 premium offsets some of the opportunity loss, so for 10,000 shares, $70,000. Your opportunity cost or potential loss is $70,000, and there it is, $70,000. So once again, none of the use cases they outlined in this video highlighting 0's potential were really necessary. I did not go into this video planning on essentially showing that the 01 models weren't necessary. I really didn't. But now that we're here, I don't know what that actually means. I think it definitely reinforces what I've been saying for a long time, where the vast majority of use cases do not need this level of logic and reasoning. The ones that do, yes, are still very valuable, but if each use case that they just showed off can be replicated with GPT-40, why would you pay the extra money for 01 mini? Why would you use 01 mini? Now yes

on my benchmarks, it blew it out of the water. Compared to when I tested GPT-4 for the first time a few months ago, much, much better. But GPT-4 has been getting better, so maybe they're equivalent now, at least for, again, the vast majority of use cases. All right, so that's most of the video. They close the loop on a couple of things, show off a few customer examples at the end, but that's the majority of the video. And it really seems that GPT-4 was not necessary for any of those use cases. Now, GPT-4 is still awesome. It really unlocks a new scaling mechanism for LLMs, inference time, or test time compute, which is fascinating and really just allows a whole other lever of scaling. But GPT-4 is plenty capable for probably everything that you or I need to do. If you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

There's Something Weird About ChatGPT o1 Use Cases...

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

There's Something Weird About ChatGPT o1 Use Cases...

There's Something Weird About ChatGPT o1 Use Cases...

스크립트