This Is The Best AI Tool Ever. I’m Not Kidding.

Okay, so OpenAI came out with Deep Research and I think this might be the most significant release over the last year. And oh boy. Look, I'm aware that I say this sometimes, maybe a bit too often, but they keep escalating the features. I think for most people this one will be even more useful than Operator as it's literally an AI agent that goes out there, searches the entire web and comes back with a report for you. And it's the first full implementation of O3. So this really, really is a significant release. And it's not just that it looks good, it performs extremely. Me and the team have been playing with this. Matter of fact, we Already maxed out OpenAI Pro account on the very first day of this coming out. And in this video I want to briefly show you what this is about, how to use this, and what use cases you can expect out of this OpenAI's deep research. Let's begin. So, first things first, this is their brand new feature that they only shipped the Pro plan as of now. Yes, that is the $200 plan, unfortunately. And even on that plan that is unlimited. In most other functionalities you only get 100 deep researches a month. The good news is that Sam Altman on X already pointed out that they will be expanding this to the plus tier and that will come with something like 10 deep researchers a month. That is in the future though, and he also pointed out that they even want to bring that to the free tier. But what we're looking at right here is the full version of Deep Research. And here's the key component to this function. This thing uses O3 full under the hood, not O3 mini, not O3 mini high O3 full. And the way you use it is like a basic GPT4O prompt. All you have to do is enable this Deep Research button and if you give it any sort of prompt prompt, it will come at you with several follow up questions. And once you answer those, you can simply separate them by commas or do it in a bullet point list. You can send that message and it will begin its Deep Research. So in this first step it only uses 4o and then it engages this Deep Research agent that uses O3 and a variety of tools under the hood. So this is not just a reasoning model, this is really a reasoning model with access to the Internet. So it has something like a Bing search engine at its disposal. And it also has the ability to write and execute Python code to organize all of that data. And a combination of those in a reasoning model will Work through your problem. As you can see it right here in the activity, it will find various sources on many of these reports that I created, it finds dozens of sources for you and then it compiles all of that with the power of the most potent reasoning model in the world, which is O3 as of right now, into a very long report for you. Now, let me tell you, that doesn't just sound interesting, it also really is. This O3 reasoning combined with these tools is really the first agent that is very accessible to everybody who can afford the $200 price tag. And the way I came to think of this in my first days of the usage is that running a prompt through ChatGPT is similar to a Google search, especially when you enable the search feature. You can also think of this as giving an intelligent assistant that is very well read, a computer with access to the Internet and some instructions, and then they have five minutes to complete the task. And that's super useful. That's why all of us have been using ChatGPT. Deep research, on the other hand, when you switch over to this, it's also like handing off a task to an assistant. But this assistant isn't just well read, it's also a very thoughtful assistant that is extremely good at planning and critical thinking. And this assistant is not going to spend five minutes on your task. He's going to spend three hours on it, maybe even more in some cases. So let's think about that. If I gave an assistant a task like this, like, hey, research all the different YouTube cameras for me, sure they would follow up with these questions, I would answer them. But then if you've ever done some product research on the Internet, you might know what the process usually looks like. You go out there, you find different blog posts, different YouTube videos, you find different comparisons and tables, you find all the opinions you can, you consume all of that content and then if you want to go to extra mile, you create some sort of Excel sheet that creates an overview. But this takes some time. I mean, think about this. If you want to consider 21 sources in your report, 21 different articles, and you actually want to read them and consider all the information in there, just finding those sources will take you 15 minutes. And then reading through them will take maybe another 45 minutes at least. And then you have to make sense of it. In this case, with deep research, it took me the time that I needed to to write this simple little prompt. I had to follow up and then I suppose it costed me 100 of the pro plan which would be $2 if you want to calculate it like that. Of course there's many other benefits there, but let's just do that. So it cost me like a minute and $2 versus well, I think it'd be fair to say that this might be a two hour research task. And I picked this one specifically because I'm actually very well educated on all the different Sony models for YouTube production. And I gotta say these are excellent recommendations. Oh, I wanted to say it's missing the A7 IV but no, it's actually right here. And look, it doesn't just end with the cameras, it also step further. Obviously if you want the interchangeable lens camera, you will also need a lens. And let me tell you, this is my YouTube recommendation for so many people. The Sigma 16 millimeter with an APS C size sensor just works like a charm. In my case I use something a bit more zoomed in, something around the 40 millimeter range, which would be this one on a APS C or this one is an excellent recommendation too. This one works too. Let me tell you, I have domain expertise on this stuff and this is an amazing report. I couldn't do a better job if I spent two hours researching this for a friend of mine that asked me about this this and again all it took me was about a minute of time and $2 to save two hours on this task. Look, I'm trying to look at this neutrally, but this really feels like an unlock and an advantage over everyone who does not have access to a tool like this. ChatGPT is already an advantage, but having this deep research thing at your disposal because it doesn't just end with product research. Let's talk about some other use cases that pointed out here this combination of the world's best thinking model, Internet search and running Python code is just something that is a real, real time save. Maybe even more so than operator for most people right now. I'm of thinking about that one right now but at this point I want to give you some usage tips because the neat thing about this is that it uses GPT4O to engage this. Meaning you can engage deep research but you can also upload files, images. This is the latest one from the O3 mini release video. So I could do something like attach an image of the different benchmarks that I pulled together and showed you in that video and then I could tell it to research the Internet and look at different user reviews and opinions on O3 mini deepseek R1 Gemini thinking 2.0 and 01 Pro and instead of me going ahead and scouring the Internet for all of this information, I can, I can just send this and all of a sudden I'm using an attached file image in this case in combination with text. I give it some follow up prompts and then it's going to run it. So you have this essential feature that I know that so many of you value, which is attaching files and working with them. Like look at this. It just took 10 minutes to complete this research. It used 26 sources. And here is a comparison of O3 mini, Deepseek, R1, Google Gemini 2.0 and O1 Pro for business use cases. And then look at this thing. This is no joke. Everything is cross referenced with the different articles where it's pulled information from. I'm still scrolling by the way. Talks about the benchmarks, it talks about users, reviews and feedback and so much more. And the wonderful thing about pulling in this many sources is that it combats the number one problem with AI outputs. The fact that they're super general, not personalized, not opinionated, pulling in Reddit, opinions on O3 mini. Custom Prompt Engineering. Here is the antidote to generic AI outputs. So that's big. And let me show you some parts of this blog post that I want to highlight that they published with this. First of all, there's this one that everybody keeps talking about, which is humanity's last exam. This is evaluation that includes some of the hardest questions you could imagine. Matter of fact, they even ran a competition. We highlighted that on the channel back in the day where they asked all the people like give us your hardest problems, will include it in humanities last exam. So basically these are supposed to be the hardest expert level questions that you can come up with. And if you look at the performance of different models on this particular benchmark, GPT4 all scores 3.3%. Something like Deepseek R1 and 01 scored 9.4 and 9.1%. O3 mini that we got last week on the high setting scores 13% and OpenAI deep research scores 26%. That's more than twice the score of O3 mini without this tools. This is the power of agents in a benchmark. Look at that. You give it the normal model, you only get 3% accuracy. And if you give it the best thinking model with access to the Internet and the ability to run code, you get 26% accuracy. That's 13 times more. And think about the fact that this is in the span of a year, maybe two years, if you measure it from GPT4O. This is what progress looks like. There's one more graph that I want to highlight here and that's this one. That's the pass rate on expert level tasks by estimated hours. And the interesting finding here is this. Look at that. So these are different tasks that would take a human, let's say 1 to 3 hours, 4 to 6 hours, or in this case 10 plus hours. The interesting thing here is that the tool actually performs better on the tasks that would take a human 10 plus hours than on the ones that would take a human four to six hours. In other words, what this model finds really hard and challenging is not the same thing that a human finds time consuming. Meaning that this is an amazing steroid for already productive people to save time, if you know what to ask it. And that's where we get to the last part of this video which are the use cases of this thing because as you might know, we love to talk about that on this channel. Matter of fact, somebody introduced a little drinking game in a top comment in a news you can use video last year where he said I played a little game here where every time Igor says use cases I drank a shot and now I'm in the hospital. And the man has a point. I use that word a ton. But I think that is really the thing that matters most with AI models for consumers like us. It's not the benchmarks, it's not all the gossip around it, it's how can this improve my life? So let's talk about that. And the amazing thing about this tool versus all other releases recently, including Operator, is they had a plethora of examples in their blog post of what you could use this for. And not just expert level examples like chemistry, linguistics or healthcare use cases. This particular tool is way more suited towards tasks that a way broader audience might encounter in their everyday to do list, like doing research for shopping. If you're looking for perfect snowboards, here's a prompt for that and here's the deep research example for it. The business use cases like analyzing adoption rates or compiling reports on any sort of data that you might want to gather. Medical research on topics that you might not be familiar with, general knowledge queries that might not be answered by Google search because it really takes multiple sources and thinking to figure out the answer. There's really a lot you can do here. And this just launched and it's behind the big paid wall. I understand that that's a big prohibitive thing for many people. But the good news is it will be coming to the plus users. And also the good thing is that you have people like me who look at these things for you. So as you might imagine, we already started working on a big use case focused video. As you might also know on this channel, we do like to take our time with those. We don't like to rush those out because I'd rather have 10 really compelling use cases rather than something that is uploaded within 24 hours. And then a lot of them are just copy pasted things from all across the Internet. So we'll be definitely creating a separate video just focusing on the use cases. But let me just outline a few things that I threw at this already where it did extremely well. Matter of fact, from everything that I threw at it, I wasn't disappointed a single time. Now, given I am a fan of this technology and I am a fan of OpenAI and what they're shipping, I understand that there's a lot of drama and controversy, but I'm just looking at the tools that are given to us and I'm thankful for that. And on this channel I try to reshare the things that I find useful for myself with you. And that's my entire angle here. I try to stay out of the politics of it all. So what did I use this for so far? Well, first off, I tried to replace some of the tasks that I'm doing on an everyday basis. Like I just showed you, tool research and comparison and research on different user opinions across the Internet. It's really good at that. Secondly, the shopping advisory use case is just incredible where you send it out to look at 20 different sites and compile all the information into one report like the one we just looked at with the Sony cameras. Okay, what else did we try? Well, I'm playing around with a workflow where I want to pull together a daily news report for myself on all my focus topics. It includes all my custom instructions, my interests, my goals, the community roadmap, things like that. Still developing that, but that looks really promising. And also what multiple team members did at this point was create a custom ancestral research study providing the names of various family members and then deep research actually goes in. And in my case it pulled together things like the Slovakia Church and Synagogue where books from 1592 to 1933 or the Ellis Island Statue of Liberty foundation passenger search database for immigrants to New York City in 1892-1957 because it found records of my family name in there. Like this stuff doesn't just pull together Wikipedia articles, it also uses those, but the name is actually well picked here because this is deep research. And yeah, sure, at this point let's mention that Google released a similar product to this without a reasoning model about two months ago. Been using that here and there, but it doesn't even come close to what this thing does because again, this feels like a very intelligent and well read assistant working for you for multiple hours and all it takes is one prompt and a follow up prompt. So I think this is an extremely promising product category. This is the first really potent version of it that we're seeing. And let me go on the record here by saying I think this is the very first agentic product that if this was free it would go giga viral Just like ChatGPT did around two years ago. All right, that's all I got for you now. Subscribe for more videos on this and I'll see you soon.