OpenAI Just Upgraded All Of Their Products

Okay, so OpenAI's Dev Day just happened.

Welcome to our first ever OpenAI Dev Day.

And this is their biggest keynote of this year apparently.

So we've got some great stuff to.

Announce today because what they announced goes beyond what anybody has been expecting. And trust me, there were a lot of rumors out there. But even though the rumors were quite large, they managed to over deliver. ChatGPT got way better in many ways, including a brand new GPT Store.

We're going to launch the GPT Store.

We where people can build their own bots. Then they upgraded their GPT4 API which is huge even for non coders because this model will also be used in the ChatGPT web interface. And then they announce assistance.

You'll just ask computer for what you need and it'll do all of these tasks for you.

Which is essentially OpenAI native agents, they don't call them that yet.

We're thrilled to introduce GPTs, but they're.

More potent than some of the agents that are out there today. So let's talk about all of this, what what it means for you as a user and then let's briefly look at what it means for the AI space because even before this conference they were ahead of their competition but apparently they didn't consider that to be good enough. And they released all of this today, starting with GPT4 Turbo. And I think it makes sense that we start the conversation with this one because it's going to be the new underlying model under GPT4 when you use ChatGPT in the web interface. Big news that some people shared on Twitter over the last week was that it actually got updated. It's not cut off in September 21st anymore.

We are just as annoyed as all of you, probably more that GPT4's knowledge about the world ended in 2021.

As of now it's cut off in April 23rd.

GPT4 Turbo has knowledge about the world up to April of 2023.

So yeah, you can just go into ChatGPT and it's up to date and they even announced that they're going to keep updating. It's something they haven't been doing up until now. So this data is going to keep coming closer and closer to today's date over time. Beyond that there's a whole list of developer focused improvements like a massive context window of 128k tokens.

That's 300 pages of a standard book.

This makes it even larger than Claude and means it can effectively take in a Harry Potter book. And all that with apparently greater accuracy.

And in addition to longer context length, you'll notice that the model is much more accurate over a long context.

And again, these are updates to the API. So this won't be available in the web interface as of today, but I think it's still interesting. So let's cover the other points here, because in my opinion a huge thing they're coming out with is a so called copyright shield.

Copyright shield means that we will step in and defend our customers customers and pay the costs incurred if you face legal claims around copyright infringement. And this applies both to ChatGPT Enterprise and the API.

So this protects all the companies working with the GPT4 API from getting sued for copyright infringement. OpenAI is going to take the hit for you if any lawsuit should come up. That's pretty huge. Oh, and talking about big improvements, I'm.

Super excited to announce that we worked really hard on this and GPT4 Turbo a better model. It's considerably cheaper.

Well, they made the whole thing two to three times cheaper to use. GPT4 was pretty pricey. So recently during product development, we had to generate tens of thousands of prompts and this ran up the bill very quickly. Legitimately. $100 evaporate like this, now it's going to be up to three times cheaper to do that. Oh, and the last but definitely not least point here is the releasing APIs for Dall E3, the vision module and voice.

Dall E3, GPT4 Turbo with vision and the new text to speech model are all going into the API today.

So you're going to be able to generate high quality images with text in them. You're going to be able to read images. So a whole new set of applications is going to pop up. I'm guessing there's going to be these wearable cameras that are the GPT4 Vision API and they'll be able to take photos of the world, analyze them and then run software on top of that. Like that is pretty crazy and something I personally am very excited for. Can't wait to see what people come up with here. But all of this is available now, right? People can build on top of it. So these are the Updates to the GPT4 Turbo API and the reason they started with it. And I also wanted to cover it first because this is really going to bring the biggest real world upgrades to apps from today on out. So over time we'll be covering these on the channel. But what might be more relevant to most people watching this video is what they're bringing to the ChatGPT web interface because they're changing a lot. Now, you might have heard this news that they're bringing all the models together. You're not going to have code interpreter GPT4, Dall E3, and the choice of GPT 3.5. No, they're merging it all into one thing, so you don't have to pick a model anymore. If you have ChatGPT plus, you're just going to log in and everything's going to be there. If you ask it to do a task, it's going to decide by itself which tools are the right ones for the task you gave it.

ChatGPT will just know what to use and when you need it.

And while that's all well and good, the big news here is they're bringing out a whole new concept. So. So take a deep breath and try to wrap your head around this, because what they're coming out with, they're calling GPTs. And what they essentially introduced here are custom chatbots that you can create yourself with natural language. So in the new GPT maker, you're just going to be able to type what you're looking for, drag and drop a PDF you want it to know about, hit the little checkbox that says, Use Dall E3 in my custom chatbot, and it's going to do all that and so much more. And the best part is, once you set this up in exactly the way that you want it to function, you don't have to publish it. You can set it to private and use it just for yourself or your organization. But they're not stopping there. They're going a step beyond, and they're creating a whole marketplace where you're going to be able to share these GPTs. You're even going to be able to sell them if they're good enough. So if you train them on specific transcripts or give them specific knowledge that only you might have, and the resulting chatbot works so well that other people keep using it well, you're going to be making money off of that. And yes, effectively, you'll be entering into a revenue share agreement with OpenAI. But that is absolutely incredible. You get to build a chatbot with their tool on their website, sell it through their marketplace and reap the benefits. So if you have some type of unique data, well, now might be your time to shine, build your own GPT and offer it to the public. I love that. So they briefly showed this store, and one example that I caught on, there was a custom GPT that was just there to create slides. So if you're creating PowerPoint presentations, no need to open up GPT4 and stumble your way through the entire process. There's going to be a customized GPT that does one thing and one thing only, create PowerPoint slides. Right? The same thing goes for every coding language and many other use cases, which we'll have to discover here together over the coming days and weeks. Now, you bet that I'll be live streaming my experiences with building some of these, so tune in for that. And yeah, as mentioned, all of these GPTs have custom knowledge bases, so you can upload any documents to them and you don't have to worry about technical topics like how is it going to be chunked and is it going to retrieve the various chunks within the document correctly, or is it just going to mess up along the way? These are all problems that people who are building custom chatbots were facing on a daily basis. None of that anymore. You just speak to it and drag and drop your files. I love that I'll be using it and I hope a lot of you will join me on that journey. And if you thought this was amazing already, well, we're not done yet. Because they're bringing auto GPTs to their own playground. Because OpenAI is taking what we this far called auto GPTs and they're giving it a very own API. They're calling this the assistant API. And what this essentially is is autonomous agents. But they're not calling it that because in their eyes, and in mine too, it's not there yet. But rather than talking about it, let me just show it to you because it's available already, you can go to platform.OpenAI.com assistance and then with your API key, which by the way, if you're not aware you're paying for every single request, so just be careful with that. You can and should set up usage limits just in case. But inside of this playground, you get to now create assistants. So let's see, let's just hit create. You get to name it, then you get to give it instructions. And this is something like the custom instructions feature inside of ChatGPT. So you can just tell it you are a helpful assistant. And this would be the default setting of ChatGPT. Or you could go into great detail and give it rich context. And at this point I just have to mention the product that we've been developing for the past four months because it does exactly this. It gives you a super rich character that you can use inside of ChatGPT's custom instructions or inside of any chatbot just like this. So if I wanted this assistant to take on the role of art director, I would just go in here, pick the art director, copy the entire character with all the reach details in here, and then just paste it in the instructions and there you go. And now we have art director assistant with all the details that are fleshed out here. By the way, this product also comes with 20,000 prompts that are specific to all the characters. We're going to be updating it to a thousand characters and you get a course, a website, a notion template and a prompt generator that you can use on any of these characters. So it's really the ultimate suit to build assistants like this. But in this case we're just going to be going through this test. And the best part is they already made this brand new model with 128K token context window available. You just have to pick the GPT4 1106 preview model. Now you can do things like this where you feed it super long instructions. And here at the bottom is the best part. This wasn't easy to do up until now. You could just activate either retrieval or the code interpreter, or you could even add custom functions with code here. But what these do is code interpreter allows you to run Python code inside of a sandbox. That means that this assistant is actually going to be able to execute code. It's going to be able to create graphs, it's going to be able to run data analysis, it's going to be able to process photo, video, all the things that you can do with Code interpreter. This assistant will be able to do by itself now. And retrieval is one of the most convenient functions because what this allows you to do is you can simply upload, upload files. And look, now I can do something like this where I just Google Nike design manual. Let's add PDF. Now what I can do is I can upload this Nike design guide inside of my new assistant here. And now I have an art director character that is aware of all of Nike's design guidelines. And that's it. This took me about two minutes while explaining it. Right, so in this case I'll just turn off the code interpreter because I don't really need it for this assistant and I'll simply hit save. And now I get to interact with this brand new assistant that at any time I could give access to the code interpreter and access to more PDFs just like this one, all in seconds. And it's using all the perks of the GPT4 Turbo model we just talked about in the beginning of the video. Oh, and by the way, they mentioned that you're going to have voice in and output with these two. So you're going to be able to create your very own assistants, talk to them and they're going to be talking back.

I am Jarvis, a virtual artificial intelligence.

And if you want them to have more knowledge, you upload more files. If you want to customize their personality or how they behave, just customize the instructions in here. This all just became really simple, even for non coders. And if you go to platform.OpenAI.com playground link will be in the description. You can actually access the brand new GPT4 Turbo model with the 128k context window today. Look, this is it right here. You just send your message in here. Oh, and by the way, for all ChatGPT plus users, if you go to GPT4 and ask it right now, when is your knowledge cutoff? It will tell you, my knowledge is up to date until April 2023. Meaning we have GPT4 Turbo inside of ChatGPT already. But now it's time to sum all of this up. So what does this mean? There's a lot of new updates. They overhauled their most important products, right? Their flagship API just got so much better, cheaper, more up to date. ChatGPT plus users now have one model that does it all. Plus there's going to be these custom GPTs that are really good at one specific thing and they're adding a store for it. Plus, if that's not enough, you're going to have these assistants that are going to just run by themselves. And the story of the day here is that all of these developments point towards one thing. They gave us the base models, then they let the developer community figure out all the best use cases, and then they just build themselves and include it as a feature of their platform. These assistants were AutoGPT, Baby AGI and all the other apps that were built around this concept of autonomous agents a few months back. The GPTs in the store are all the various AI wrappers, right? You have one that is really good for creating slides, one that is really good to help you train your dog. Now, from today on out, most of those are obsolete because you have them right inside of ChatGPT. And you can expect the same to happen with all the new APIs. People are going to be building vision API apps, and then in a few months they're probably going to come out with versions of the best ones that they create themselves. And what this nets out to is that it's quite bad for the big players that have put a lot of money into developing their very own assistants. Whereas now people can just do this in a few clicks. But who it's really good for is the users. It's the people who use it every single day. You can now build your own assistants. You can now build your own GPTs. It's cheaper than ever. It's easier than ever. And the next round of innovation is going to start today with the release of their Vision API. I personally can't wait to see what people build with it. If you care to learn more about the best use cases of the Vision API, which is probably the next big chapter of OpenAI story, then check out this video where I dissected a research paper that showed off over 100 vision use cases. I'll see you there.