The Biggest Announcements of OpenAI's Dev Day

Well, well, well. There we go. Nice. Welcome, everybody. Yeah, I've been digesting what just happened. Hello, everyone. Hello, hello, Hello. So let's see, can we get the music working in here? First time streaming. Streaming mobile. Incredible livestream. So what do you guys think? Look, we're gonna get into all of this, right? We're gonna talk about each part of the presentation. I mean, they effectively announced agents, they announced the ChatGPT marketplace, and J announced the GPT4 Turbo API that comes with improvements. Improvements. But not even the most curious and open minded rumor hunters on Twitter found out, right? So what do you guys think of all this? Which I would be curious. Like which one of these sites do you guys the most? And we're gonna talk about them one by one. And what I'm excited about is that on the stream we're gonna get to talk to them as a community. Right. So we're gonna go point by point and I'm super curious to hear what you gu. Right. It's not just gonna be about my thoughts while I share those, but. Okay, okay, okay. Yeah, let's get the music out of here. I don't have my mix board with me. Good. So is it better now? Is it better with the music? So what I plan on doing today is I want to have an open discussion with you guys with the entire chat here, what you guys make of these various announcements, because this is not just about what I think of these. Right. I'll certainly share that. But I really want to hear what excites you, what you're looking to build, what you're looking to try. I mean, we're getting a ChatGPT app store, right, where you're going to be able to build your own bots and share them with the world. How cool is that? They essentially announced the API for their agents. Yeah, there you go. Music is fine now, right? Like, it's gone now. I just turned it off. I think that's fine. We don't need no music. My music is too loud. All right, all right. But I turned it off, my friend. I turned it off. So let's see. Cool. So it should be. It should be fine now. Yeah. So we'll talk about all this. So you know, I took notes on everything. I'll summarize everything for you step by step and then we'll just talk about this as a community. I can't. I can't wait to hear what you guys think because I certainly, I certainly have my thoughts here. Okay. Okay. So we'll just make sure to get the audio right here. By the way, this is the first time that I'm streaming from abroad. I was a. You know, I had. I had my concerns here, but we'll figure this out. Voice should be fine now, right? Video looks good as always, and it's cool now. All right, so let's talk about this. You know, let's. Let's just begin by talking about what we heard. I mean, I'm guessing all of you have been participating in the live stream that just happened, and there's a lot to talk about. So first of all, good to have all of you here as per usual. I just. I'm sorry. I just love doing this so much. Please share the country that you're in and chat right now. It's always so fun to see how international this community is. It couldn't be more global from my experience. So just share where you're from and. Yeah, I'll maybe start. I'm in Portugal right now because, as you might have heard, as you might have seen, I kind of moved countries. And what I did is I, you know, I brought the whole setup. Like, here's the studio mic. Here's the massive lights. Here's. Here's. You know, I'm not cutting corners. We're doing hair, light, everything out here. So where are you from? Question. I'm not aware what's coming. Good, because then you're in the right place. I'm going to summarize it for you, but first, let's look at where all these lovely people are from, because this is truly crazy. Czech Republic. Good to have you here, man. Nice. I'm from. I have a Slovakian passport, so we're neighbors. Denmark, United States, Germany, Louisiana, usa. India, France. Guys, some of you stayed up all night for this, right? Texas, Berlin, Poland. Man, I'm absolutely loving this. California. Austria. Nice, man. That's where I grew up. Yeah. Paraguay. England. All right, all right, all right. So that's. That's absolutely amazing. Zion, Leonard from Milan. Berlin. Guys, this. This community couldn't be any. Any more international. Absolutely love that. So let's get. Let's get into this step by step. Let's talk about what they announced. Okay, so first, the broad overview. The broad overview is they essentially had three big topics, right? So the first one was the Update to the GPT4 API. They made it, you know, they call it Turbo now. It's cheaper, have a massive context window. I mean, they just had to, like, kind of top Claude, right? Claude did 100K. Everyone was like, oh, Claude is Context king. It might be not as good, but they have the long context. So OpenAI, you know, had to go one level above and be like 128k, man. So they did that. Then they released the ChatGPT updates with GPTs and the GPT Store, they call that. Yeah, the GPT Store. Exactly. Where you're going to be able to build chatbots with natural language, which means just like this, you're going to be like, hey, please do this. And could you, you know, have this type of personality? Oh, and here's a PDF from my desktop that you'll just like drag and drop into it and it does all the technicalities for you. Right? No more. Bet like anybody who has built chatbots will know how annoying a lot of the technical parts are, right? Like, is it chunked correctly and then is it retrieving the chunks correctly? And then you kind of have to go in and you have to structure your data correctly. No more of that. You just kind of like drag and drop it in. And yeah, again, like, they dropped. They dropped a bit of a. You know, they just. Again, as. As per usual with most OpenAI updates, they just deleted another, like 10,000, like little OpenAI wrappers that people built. But. But that's how it goes, I guess. So they're gonna have to GPT store. The polls in here are. Yeah, a lot of Polish people in here absolutely love it. Yeah, there you go. I like it. So, yeah, they're gonna have the GPT store and you're gonna build your own custom chatbot. So let's look at that and maybe let's even go into the presentation and kind of. I know there was one screenshot in there where they showed the store. Let's talk about that. They showed some, like, examples in there. So some of them are for specific coding languages, which is really cool, but some others are for specific use cases. Right. So just like you had these AI apps that were really good at building a PowerPoint presentation, right? That was like the one use case for the app. Now you're going to have the same thing, but inside of ChatGPT, you're just going to go into the store, you're going to like, use the GPT that is there for creating PowerPoint slides. That's one of the examples that I saw on screen there. So I think that's really cool. Bots will replace plugins. Yeah, that's essentially what's happening. They kind of. They didn't have a single mention of plugins. Right. So these new. Yeah, these new bots, these new GPTs are essentially like have plugins built into them. And guys, this is still just overview. This is not even the in depth look. Right, so next up we're going to be talking about Assist, the Assistance API. This. Okay, so in one sentence, essentially they're doing agents, right? But they're acknowledging the fact that agents might not be there yet and therefore they're not calling it, you know, agents API, but they're calling it Assistance API. So I think that's fair because, because the whole, you know, auto GPT movement, I think everybody agrees at this point that it was kind of like overstated. I think the idea is something that everybody grasps in like a second. It's like very catchy. But then in practice you're just like, okay, this thing is like interesting, but it doesn't actually get anything done effectively. And then people kind of like abandon it after playing with it a little bit. Right? So they're kind of acknowledging that. And they're doing their own version, which they're calling assistants. There's function calling and code interpreter and retrieval, and the ability for it to have long, coherent conversations all in one. You're going to be able to use voice to input output. Yeah, AutoGPT was just too early. I think that's accurate. Daniel. So it's all coming together, I think. Yeah. Robert. I think that's kind of the perfect summary. Like everything we've been following this year, all the various announcements, you know, like every time a little like, announcement came out, I made a video about it. Like, oh, nice. Now we have Bing browsing. Now we have plugins, now we have this, now we have that. And what this event was really about was about like bringing it together, giving you one API where you can access all of it, giving you, you know, one chatbot where you can activate all of it. If you have ChatGPT plus, you're just going to open it up and yeah, it's going to be one model. It's going to be GPT4 with everything enabled. And when you ask it, you know, to create you a diagram, it's just going to do it. You're not going to have to go into the App Store and look for a specific plugin, activate that, and then you're going to like, prompt that you want a diagram and it's going to kind of, this was my experience a lot of the times it's going to kind of not even use the plugin that you just activated because you didn't prompt correctly for the plugin. Because you maybe didn't mention one of the functions of the plugin. So then you have to go in and use a different prompt to find out what functions the plugin has so you can call them. And like now four minutes passed and you're just trying to create a diagram. If it would have been faster to just open up, paint and do it yourself, right? No more of that anymore. You're just gonna go in, you're gonna be like, hey, I want a diagram that is similar to this. And then you snap something with your phone and it does that. I think that's really cool. And yeah, it's all coming together, is accurate. I'm confused. I have the plus plan that doesn't apply. I believe it's unlimited. So this is a thing that I can't answer yet because I'm not fully clear on how much of this is coming to the plus plan. To me it seems like the GPTs are going to be kind of a add on extra thing. So if the plus plan, I think you're going to be getting, you know, code interpreter, GPT4, the new Turbo model, you're going to be getting daily free, you're going to be getting voice input, output, vision, all that. But when it comes to the chatbots, because they mentioned the revenue share model, to me, it seems like it's going to be extra thing, right? And just given by the fact that you're going to be able to build your own and kind of like sell them and make money off of it, has to mean that it will cost extra, right? They will not be giving you a part of the ChatGPT subscription. So yeah. And also they updated the knowledge to April 2023, so fantastic. So let's talk about all this and let's define agent along the way. Let's answer all your questions in here. Ireland in the house too. Nice. Good to see. So I'll be reading comments here throughout, so you can feel free to chime in at any point in time. I think some of these points that we'll be covering are incredibly interesting to discuss and I'm just excited that we get to do this in a collaborative manner like this because some of you certainly have viewpoints that I'd be interested in hearing and you certainly have a lot to add from what I know, because the average IQ in this community is way above average, let me just say that. So will this be something that they were calling Auto GPT? No. So they didn't use that terminology. But essentially the assistant API is very similar to the auto GPT concept. Yeah, they call it assistants and essentially it can run code by itself. It can search the Internet, it can. It can call functions, you know, so searching the Internet is just one of the functions it can do. Any thoughts on grok? I think that that's kind of a separate topic that I think it's exciting. I think it's cool. I personally kind of love the unfiltered models. And you know, the first channel, the first video on the channel was actually me showing some of, you know, the crazy illegal use cases of GPT 3.5. Because, like I started using it on December 1st, which was like one day after release. And that was obviously, that was one of the first things that I tried out. I was like, what? Like, can it do these things? And then, then I just created a video about it that did well. And yeah, here we are a year later, me doing this full time. But yeah, I think Grok is exciting. I signed up for the waitlist, by the way. If. Yeah, let's just wrap on that. You can, you can go on the GROK announcement and at the bottom there's a link where you can sign up to the waitlist. It's kind of hidden, but it's just like one button and you can get it. Anyway, I'm excited for all tool GPT4 default mode and Turbo model coming from Canada. Random bleach fan. Much appreciated, man. The super chat in the house. It's certainly appreciated. Good to have you here. Yeah, I love it. Super international in here. I'm excited for that too. I think that's going to be huge. I can't wait. I have like the standard set of like 20 use cases that I keep returning to. And with this update yet again, first thing I'll do is I'll just go through them and I'll run them through the new model that combines everything into one and I'll see what it does. Maybe it'll start generating images when I'm asking it to. Like, I have one where I like mind map, like a decision tree. Does that make sense? Yeah, I think. Yeah, the prompt is something. I don't know it by heart. It's kind of long. But I create like a mind map of like various decisions that I could take around the piece of content. I think it might just start using Dall E for that. Right. I haven't tried that yet, so that's gonna be huge. Vision API is gonna be huge too. So. Okay, so let's put a little bit of organization on all of this. And let's start covering this one by one and I want to hear your comments. Okay? So, you know, get your keyboards out, get your, you know, warm up your thumbs because I want to hear what you guys think of these. And then in the end we're gonna do a little poll on which one is actually the biggest one, because I don't know, I have some favorites here, but it's not, it's certainly not clear. So let's start by talking about GPT4 Turbo. Okay, so, you know, let's go from a place. And now it's interesting, you know, we could talk from a development angle where I kind of like assume that you have a bit of development basics, but I don't think that's the best angle to take here. I think we're going to go from an angle of like, you've been following this channel a little bit, you've seen a few videos, meaning that you probably know what an API is already. Right? I'll assume that, um, but mostly I'll assume that you're like an avid ChatGPT web interface user that is interested in these developments. I think that's the angle I'll take here. Okay, so GPT4 Turbo, what's up with that? So this is a, it's a brand new API that they're going to be offering and essentially it's similar, but even they added even more than they did with the upgrade from GPT 3.5 to 3.5 Turbo. So one thing is the massive context window, okay, so they upgraded it to 128k tokens, which is, you know, into Claude. You can, you can feed, you can feed a Harry Potter book into Claude. So now you could even feed the longer Harry Potter Potter books into Claude. I think Sam said during the presentation, 300 page book, but I think that would need to be large pages. If you're looking at like a pocket format, then 128k context should total somewhere around 100k words, if I'm not mistaken. And I'm just like thinking here on the fly, but basically they're going to be the biggest, this is going to be the biggest context window we have access to out of all the various large language models. So right now it was Claude with 100k and yeah, they just had to one up it with 128k. Right. So I think that's really exciting because one of the biggest limitations is when you're having long form conversations with the, I'm going to call it, what do you want to Call it with the chatbot, with the. I don't know, I don't think I like the term AI, but if you're having long for conversations, it just, especially in the web interface, it forgets so fast, right? You can't, like, as you know, if you've been following the channel, as you know, one of the most powerful techniques you have is multishot prompting, which means you give it multiple examples of something. So if you're giving it multiple examples of like a long blog post, then you know, in the middle of providing like your four or five examples, it's just going to run out of context. Not anymore, because you're going to have this massive, massive context window. Plus they also mentioned it's going to have a higher accuracy, which is actually big because one thing is like, oh, here we have this massive context window. But then it kind of doesn't really respect what's within the context. So I really enjoyed hearing that because that has been my experience with Claude. Sometimes we just feed it, feed it, feed it. By the way, if you're from Europe, you need to use a VPN still, I believe, right. I think UK has it now anyway. But like you just feed it, feed it all this context, all this rich context, and then you kind of like prompt on top of that and ask for some of the info in there and it just, you know, it ignores it. So that's not going to happen anymore, hopefully. Because they said there's going to be higher accuracy. Another one is there's going to be better function calling. So this worked pretty well. But again, it was similar to like the accuracy with the long context. It just didn't. It just had these like.