In 3 Tagen eine App: ChatGPT O1 Pro für 200$ LOHNT?

ChatGPT has a new pro version for a full $200 a month and I got it for you directly at release and tested it extensively over the weekend and developed a full app, a mobile app that is currently being checked in the play store on android and then released to everyone. iOS not yet, that's coming, it takes a little longer, the two need a little bit. And yes I would say I can now review very well what OpenAI's GBT-01 Pro can and what it can't do so well.

But before that, let's look at this pricing more closely and we want to answer the question who needs it, who is worth this 01 Pro offer? Okay, so what's new? Well, we have ChatGPT, that's the normal web browser.

and there are three payment options, technically four. We'll get to that in a moment. First of all, we have the free payment option, then there is a $20 per month option.

That's the one we've known so far as a payment option. And then there's even the $25 per month option, which few people seem to know. The difference between the two is quite simple.

With the $20 option, OpenAI says, yo, we take everything you say and use it for the AI training of future models, for example the new one. With the $25 per month option, which I've had since release months, it's just said, no, your data is your data. We don't use it for training.

That means, sensitive content is, at least if you trust it, which is of course always questionable, but if you trust it, then the things are not counted into the training. So, and now new is the newest variant, the $200 per month variant. And it has it in it.

In the end, there are several models. The free users have access to GPT-4O and O1 Mini, both with access restrictions. That means, not unlimited requests are possible, but limited to a few per day.

Or per, whatever you decide right now. Sometimes it's a week, sometimes it's per day, sometimes it's per month even. But yes, in the end, both models are available.

If you pay $20 per month, it is currently the case that you now get O1. So far, by the way, it was only the O1 preview. Now the finished O1 model is out.

And the O1 model has a special feature. Just like the O1 Mini and the O1 Pro model. Because the O1 Pro model you get at the $200 per month variant.

And the difference is now a very subtle one between O1 and O1 Pro. It is basically the same model. So the same model is used in the background.

The same Large Language KI model calculates O1 and O1 Pro answers. The difference is the following. This model asks itself regularly.

That means, if you ask a question, it starts to build a structure. What do I have to do for that? Does that make sense?

What did I research? Okay, I'm questioning myself. And so on and so forth.

So, this process takes a lot of computing time. And in the end, the more computing time you give O1. That was now found out by ChatGPT.

The more computing time you give in there, the better the answers will be. We'll take a look at benchmarks in a moment. Basically, it is like this.

For 20 dollars per month users, the computing time is limited to O1. That means your model basically just doesn't think about an answer for that long. It does think about it, but just briefly.

It's like I only have half an hour for a class assignment. And the O1 Pro model is supposed to think about it until it's really done. In practice, this is about a factor of three in my computing time.

Where it's different. That means the O1 Pro model thinks about three times as long. But that's definitely not to be seen as legal.

That's a drastic difference. The O1 model, the normal one, can think for more than a minute. While the O1 Pro model can only think for a few seconds.

That always depends on your request. But if you ask the same question, at least for me, O1 Pro thinks about three times as long. Okay.

And now we want to test, or rather evaluate, for whom this unspeakable, ten times as expensive model O1 Pro is worth. Yes. You probably already have a guess, when I upload this video on this channel here and upload it with this foil layout, that it's not for the general public.

Although I have to say, yes, we'll get to that in a moment. It's actually not that uninteresting, even for normal people. But well, we want to clarify whether this thing is worth it at all.

So, now let's get to my project for a moment, so you can see what I've developed with it, or rather what I've actually tested now. And let's put it this way. Benchmarks, I'm over benchmarks.

It no longer makes sense to use my benchmarks. Even 4. 0, so the cheapest model that you can even get for free, has already really played through almost every benchmark that I could invent out of the baton.

Besides, I program everything ready benchmark. And then it had already played through O1 Preview. That's why it's ridiculous to use this benchmark now.

So what I did is the following. I wanted to develop a complete app. This is the app you can see here.

It's not a small app. It's a complete RSS Newsreader. So, that means an RSS feed.

I explain that on the other channel in great detail. I have built an RSS feed for my patrons, or even a free RSS, which in the end summarizes the most important news that have happened in the technology world. And this RSS feed, it's first of all, so maybe a very short explanation of what I've done.

That was before. I have now set up several RSS feeds for the news from the technology world. The one RSS feed you see here, the high ranking RSS feed, it's just available to everyone publicly.

You can also use it. There are the highlights from the entire technology world, from the entire technology world in it. That is, those that are really important.

So, in addition to that, I have also set up several categories, something like artificial intelligence, so AI as a category, something like hacking or cybersecurity, everything that has to do with privacy, network politics and so on, hardware, blah, blah, blah. And each of these categories, each article receives a rating in each of these categories. And based on this rating, it ends up in your feed or not.

You can use it if you are a patron, because by the way, the link in the app is even linked. And below, of course, too. But if you are a patron, then you can put it together yourself.

And then you have your RSS feed, which reflects exactly that and your interests. That means, for example, you can say, I want to know everything in the area of cybersecurity that has a rating of 6 or higher. I'm not that deep into AI, I'm enough with the 9s and 10s.

And with privacy, I want to know everything, that is, from 1 to the top, for example. You can put it together completely custom and there are now more than 10 categories. That means, I think you can say quite well, it's even over 20, I think, you can say quite well what kind of feed you want to have.

And exactly this feed, it has already been there before. So that wasn't the app. So, that's an RSS feed.

And there are several RSS readers out there. But there are a few functions that I missed. In an RSS feed you can also enter audio.

That's not the case with mine yet. I want to program that over Christmas time. That means my RSS feed will then get an audio.

And then I don't want this, that I have to scroll through. With the RSS reader that I used before, the articles were always cut out, which I have to write myself here. So I have to write a summary to an article.

And that wasn't just how I had hoped for it. And that's exactly why I developed this app here, which really gives me a reader, who shows me the article once. You can see that here.

That means, first picture, title up here. Here we have the actual article. And I just focused on that when I write these things, that I completely give up clickbait.

That I write a summary. That means all relevant information that you need, is already in there. And it's not a, hey, look what the hell is happening here.

Please click on my message so that I get more clicks or something. I explicitly threw that out there. I threw out advertising.

Then I have here when the article was published and linked the source. And then we also have a checklist. We have it here.

I can just remember things and they are available so that I can look at them later. I'll get to that. Then I have a share button here so that I can just send links somewhere else.

And then I have marked an old unread here. There is a checklist in the background which articles you have already seen. And if you just go back into the app, then you see the first article that you have not read yet.

And so you can just say by hand, hey, I haven't read the article yet, so that you can get back to it. So then if the article or if any feed that works, for example, also with podcasts, by the way, if any RSS feed contains audio, then it will or I can set in the settings that the audio is played automatically or not automatically. Then I have to click on it on purpose.

And if the audio is played automatically, I can also activate auto scroll. One of my patrons brought me on it. And the idea is the following.

If I have an audio track on each article, then I just do auto scroll and then it basically runs through. That means I can, for example, simply connect it to my car and let it run. And while driving the car, I can basically listen to the news and because it always goes automatically to the next news, it never stops.

I don't have to actively press it. One of you just brought me on it. So, in the end, these are all the functions.

But you see, it's not a completely banal app. This app is definitely a bit more complex. And I wanted to show you that you can actually really evaluate.

Can I develop something with it? The answer is obvious. I can develop something with it.

But where did it go well? Where did it go wrong? And what did I actually learn from it?

Just a quick hint. The app is soon available in the App Store. I know exactly that the question will come.

It is only not yet available for everyone. You have to do a closed beta on Android by now. I already have it for that.

I think we have about 70 tests or something. And they have now then as soon as it is reviewed. It has been written over the weekend.

Friday, Saturday, Sunday I wrote it. Sunday evening I put it in the Play Store. And since then it has been reviewed by Google.

And as soon as it has been reviewed by Google, it is available for the testers. And then I will also publish it. The link remains the same.

I will share it with you below or at the end of the video as a QR code. Then you can save it. You can just look around around Christmas time.

And then it should also be available. Or maybe at the beginning of next year. But I'll tell you again on Instagram.

And now you might ask yourself where is this RSS feed actually located? Basically the app. You can't even register there.

But the RSS feed is located with our partner Infomaniac today. The ethical server provider in Switzerland. Ethical in the sense of in Switzerland you have incredible data protection requirements.

That means they save as little data as they can and are extremely responsible for it. And because we were there we could also see what is actually being done and how exactly data is actually processed. That is the one ethical aspect.

The second is by the way, super cool office. We were also allowed to go in, it's beautiful. But the second aspect is sustainability in the sense of the environment.

Because Infomaniac has the probably most advanced data center that I have ever seen in the sense of reusing energy. That means they use the entire heat energy that comes from a data center. And they just blow it into the remote heating network of the city of Geneva.

That means they heat the city with it. And that is so cool. I will explain that in the new video that I am still building.

But it's just a game changer for everything. Plus you can of course create your own application just like in a normal cloud setting. For example open source models like Lama 3.

2 by the way the latest model. You can also use Lama 3. 3 and Mixtral 8.

22b and so on and so forth. You can just access everything at Infomaniac. There are also many other applications just take a normal VPS that I have taken for this RSS feed a server that is basically with you from 3 euros a month which is really, really cheap.

So for me just a completely logical conclusion to go there. And many, many thanks to Infomaniac for supporting this video and all of my projects which is just super awesome. That's why I can offer this RSS feed so openly.

Because if I had to pay so much for the server it wouldn't be that easy. But back to the topic. Let's take a look at the app itself.

I have developed the whole thing in Dart and Flutter. A framework to which I have already made a whole tutorial series. It's cross-platform.

Basically I can use it not only as an Android app but also as an iOS app as a website I even think as a native Linux application Windows application and MacOS application. But I have never done that with Flutter and I have never released an iOS app out of it. Which makes me a bit nervous because I still plan to do that but I don't have an Apple device and you need one for that.

So we did that with Android. We wrote it in Dart and Flutter and it became an Android app. And as you can see there are quite a few files that we have built together with a nice folder structure and everything else.

So, what did I do? First of all I used the Obon Pro model and said these are my specifications these are the features I wanted these are the ones we just discussed. The code that I have at the beginning there was none but later it became more and more and then I said this is the documentation for cases where it was a bit difficult where for example ChatGPT had the old version because something new came out I then partially just put the documentation one by one from the website or from the product page from the library page for example and said this is the docu build it for me.

In the end I got my new code from ChatGPT 01 Pro in several files I actually I had to say it a few times but I always got the whole file that means in the most cases I really just made very stupid copy-paste There were a few cases where I touched the code myself but then it was really just individual lines or just went over where an error was displayed in the IDE and then I said this error I can fix it myself for example, I just removed my const that was all I did at most I removed my const I didn't do anything else I wrote almost no lines myself But what I mainly did is I read the code and compared it with the one before that means I built a git repository and then compared what is new code in the gitdivs and what is old code and then looked where exactly are the changes although 01Pro gave me in the most cases directly and then I just entered the code and executed it and then partly just the error message again in 01Pro and then I got new answers and what I got were actually up to four full files so from those where you just saw up to four finished files completely with the whole code up to 200 lines but always very clean documented always very unfortunately extremely accurate just exactly what I wanted and the only thing I did is I gave feedback for example hey I don't think the design is so cool for example the title on the start page was previously at the very top as an address bar title I didn't think that was so pretty then I said hey, maybe we can do that at the bottom as a band so I just programmed it in English then I just entered all the error codes and said okay, do it again here is a error and I have improvement ideas, so to speak feature wishes from my side and said how I would like to use it and what to change and that worked amazingly well especially if you look at the difference between 01 so the normal 01 model and the 01Pro model there is a kind of new benchmark called OpenAI that you can use for this kind of models that can think for a long time so these reasoning models and basically this is very decisive what is written there is hey basically if you look at the benchmarks of 01 and 01Pro mode just the normal benchmarks then they are relatively close to each other what is new here is this 4 out of 4 reliability and it gives if we ask the same question in 4 out of 4 cases how often is that right or is it correct in 4 out of 4 cases and the questions can be very very complex so here a competition code Codeforces I don't know if you know Codeforces this is the benchmark that probably comes next what we do but in this case I was here in 75% of the cases with the Pro mode and in 64% of the cases with the 01 normal 4 out of 4 answers at Codeforces the correct answer was found that is competitive coding that means you have an algorithmic mostly algorithmic problem which can be partially ridiculously complex and you have to solve it and if we compare the benchmark if we just look who can do it then 01 is almost as good as 01Pro mode but if you say okay it is always correct then 01Pro mode is much better and that's what I noticed in my test when I said I want this and that as code it really had up to the finest detail in the most cases and I never had to fix the same mistake twice so something like it rubs and doesn't understand it anymore or it builds me something that wasn't right that never happened in the most cases what I said came out exactly the same which of course is not always accurate because we know exactly that you can misunderstand specifications that's what I realized myself and I try to express myself very well I divided my workflow that means it wasn't just coding from A to Z but first of all I had to understand how to set it up I called it devsetup for that I had to build an installation all the tools Android SDK, Flutter SDK I had to install then there was the project setup where we saw all the files I had to think how do I divide it logically so that it is divided into modules then came the coding and new features because I invented myself again and again if you want or that it was hyper-agile I don't know and then there were always new features and then I wanted to deploy the whole thing at the end and we look at all the steps individually because I have to say when coding that it is no longer a junior practitioner as it used to be with ChatGPT this is a software developer who knows how to develop software so literally everything I said has built the thing and sometimes came as feedback something like hey if you want to scroll like this I actually wanted to scroll horizontally with the app you have to scroll left and right because the articles sometimes did not fit depending on how big the smartphone is they are too big and you still have to scroll in the container because some people have set large writing etc.