Open AI's "OPERATOR" AI Agent - Release Date & Speculation

Hello everybody and welcome back to the channel. I think it's finally time we had a talk about OpenAI and what their plans are for the future. What can we expect? As you guys probably know, and I don't need to really explain, OpenAI is kind of on top of the AI game and has been since AI became mainstream. Sure there are some areas in which they get beaten out and right now that's video generation, image generation and some other small areas. But they always seem to lap back and become the king of whatever they do once again, every time an announcement happens. Anyways, OpenAI is preparing to launch something in early 2025. It's going to be a new AI tool called Operator. The goal here is significant advancement in automation and agentic AI based technology that going to be designed to automate complex tasks with multiple steps and complete those tasks in a timely manner with minimal human interaction. Basically. Hey Operator, order me a pizza. Hey Operator, can you search Amazon for the cheapest version of blah blah blah, whatever it might be? Basically the overarching goal with Operator is to vastly simplify complex tasks for humans. Things that normally could take you hours of research and navigation across the web, comparing different things can now be automated. You can do other things and it's just gonna come back to you with whatever you asked for. And that's pretty darn incredible. The keen amongst you will start to think, hey, there actually are some AI tools that exist that are supposed to do this. And that is very true. Most notably recently CLAUDE announced their very own computer use model. And it's, you know, alright, it's not great. It can do basic things, it can go ahead and like order you a sandwich or something like, or it could click around your computer, open apps for you. But in terms of real world use cases, I haven't seen anything yet that screams hey, I need this in my life now, because it's just not good enough. There's also the new Microsoft agents which we talked about I think last week and that seems even more promising than what Claude has brought to the table. But then again, that Microsoft agent seems to really be more like a combination of four agents working together in a loop, which isn't bad, but certain tasks might falter because of the limitation of those four specific models. I think there was an overarching task delegation, there was a web searching agent, there was a coding agent and then an execution agent. And those four can do a lot. But I think OpenAI with its operator here is going to come to the table with something entirely new. It'll be one obviously multimodal model in my estimation. And I think this model is going to have an inherent difference. I think that the architecture behind it is going to be something that is from the ground up. OpenAI research that we haven't seen before, and I think that they have probably specifically trained it on human actions in completing tasks over time. Like I said, OpenAI is known for doing this. They are, whether you like it or not, known for pioneering technologies like Chain of Thought or even just large language models themselves. ChatGPT is kind of when AI went mainstream and companies like Google Meta X all followed in their footsteps creating their own models. Now the tricky part about all of this is interacting with the computer to actually complete these tasks. There's a lot of speculation about how that might all play out. For example, Claude computer use right now just takes screenshots and then has commands set to move the mouse to this coordinate and then click or move the mouse here, click and then input these characters from the keyboard, which works fine, but it's still pretty clunky. I'm really thinking and wondering about how OpenAI might tackle this problem. Have they built their own software that runs on a Mac or Windows PC that navigates the web all by itself in a way that isn't the traditional mouse and keyboard interactions? And how does it tie in to the large language model? Obviously it's going to have to be multimodal. Honestly, if I had to take my best guess, OpenAI has somehow figured out a way to train the model directly on the computer actions. So while the model is thinking and reasoning in real time via text tokens, it can also produce tokens that translate into computer actions. Now, I'm no AI researcher or scientist here, so take that with a pretty big grain of salt, but to me that seems like the most natural and effective way to implement something like this. Like I said, I have minimal knowledge. I don't even know if this comes with some challenges or barriers that I just can't see because of my lack of knowledge. Anyways, the scheduled time of release for operator is early 2025, potentially January, which seems like very soon, which would mean that they've been working on this for a while. But again, that wouldn't be something new for OpenAI. They like to surprise us. GPT4 omni with the advanced voice mode was a surprise. When they first announced Sora, it was a very shocking surprise because it was a lot better than any video generation model we had seen prior. And even though they still all the way at the end of 2024 haven't released soar to the public. The rest of the AI space has caught up pretty quickly, which means that there's no doubt in our minds that OpenAI is very capable of delivering something that meets what Sora advertised to us. So we will have to see in early 2025 if operator lives up to our expectations from OpenAI, which are of course the highest in the industry. Many of you might have also heard that AI has recently hit a wall with remarks from top people at OpenAI saying that scaling much beyond GPT4 doesn't seem to yield much better intellig. And of course, when I say scaling, I mean the parameter count, the parameter size of the models themselves. And I think this was pretty obvious to anyone who is watching the industry. We didn't know exactly when we were going to hit a scaling wall with traditional transformer architecture, but it was obviously going to happen at some point. And now that it seems pretty clear that this is where the wall hits around, you know, GPT4ish territory, a shift in direction is absolutely needed. And I mean, over the course of this year we quite clearly, with the release of the chain of thought models which open up a new scaling vector, we have figured out where the scaling wall is with training. This is around the parameter size with the data that we have and the architecture we have to achieve the results we're looking for with current hardware. Any larger and it just doesn't make sense. But now we can scale in time. We can scale while you're using the model during inference. The longer the model thinks, potentially the better the answer we get. And right now we have access to the 01 Mini and 01 Preview models which do just this and they are a lot better. But Sam Altman was recently quoted saying that the O1 preview and O1 mini models are like the GPT2 of this scaling, which means we have a long way to go in terms of scaling in inference time. So is AI slowing down? Has AI totally hit a wall? No, not really. We hit one wall, but we've opened up a brand new door and the wall is super far away. So we've got another couple years before that wall gets hit. And then again we have yet another form of scaling, of course with Operator for this new model, which I expect to be fundamentally different. How many steps for a task can you give it? What kind of tasks will it be able to execute and how far can we push it? Those are the big questions, right? That's why I really do think that next year 2025 is going to mark a significant change in the industry. There is going to be a larger focus, especially if this operator hits home towards more autonomous agents. We've been experimen with autonomy for years now, but haven't got too far. If this is a huge leap, well, everyone's gonna know it's possible and work on it till the cows come home. And we could expect even open source computer use models in 2025, which would be insane. The real reason I think a lot of things that OpenAI does are so exciting is not necessarily because you get access to the OpenAI technology itself. That is an exciting part. But the more exciting part is the way the rest of the industry reacts and then tries to copy and catch up to OpenAI as soon as possible. And with that we get things like free models, open source models and overall cheaper costs. Competition is great for the consumer. If OpenAI and Sam Altman are to be believed about Operator, it's going to be the next big shift in AI. Now is probably the right time to consider the implications as well. First of all, let's talk about job loss. We already are seeing it with the AI technology we have today. ChatGPT, just text generation as a whole has already done crazy things, but you still need a human to at least prompt the AI, take out the text, put it somewhere else. With something like Operator, it can do probably the whole job all on its own and you'll just need someone to essentially delegate tasks. So job loss is definitely a concern. And by the way guys, just as a side note, I still have yet to see any like really convincing plans to deal with AI created job loss in the economy. If you have any ideas, if you've heard anything that seems like a good idea, let me know because I'm not really like fully convinced on anything. Even like ubi. I don't know man. Anyways, obviously there's also gonna be other concerns like privacy and security. If you've got an operator AI just controlling your computer, who says some whack job can't install it on your computer and jailbreak it to do nefarious things. Criminals with AI could be very, very dangerous. So that's something that OpenAI is typically pretty decent at dealing with anyways. And I bet this thing will be nanny down. So if you're concerned about that, I wouldn' be too concerned. At least in the early stages of 2025. When we should be concerned is when the open source models start to come. The open source computer use models which have more power potential. So more extreme good and more extreme bad outcomes are possible. This isn't the first and won't be the last time that I reiterate this point, but we have just barely dipped our toe into the water of what AI is truly capable of. Even without new expanding technologies such as operator, current AI technologies haven't even been remotely close to being fully exploited and used to their full potential. There are still new ways to prompt ChatGPT for certain use cases and scenarios that haven't been explored. And the same goes for every other modality such as art generation, music generation, image input. The list goes on. My conclusion, folks, is that the AI explosion is not slowing down. It is very real. It might take small breaks, but it never truly stops. And I still firmly, years later, believe that this is probably the largest boom in technology that the human race has ever seen. Taking into account things like electricity, taking into account things like the invention of fire back in the caveman days, it's really crazy. If the human race can push AI autonomy to its limits, what can't it do for us? And of course, that would lead us into the whole ethical discussion of AI. Can it become sentient? Can it become conscious? Would it have rights if it did? I'm not going to get into all of that because let's be real, that's for a separate video, but something to think on. Something to think on. It's really interesting because in the past three and a half, four years, we've seen literal magic happen before our eyes and us all get access to it and then get accustomed to it to the point where I don't know if I'd want to go back to living Life without having ChatGPT by my side. It's very, very useful. And yet still a lot of people doubt the power potential of AI technology and the change that it can enact on our future. I'm just saying, guys, I don't see the magic stopping anytime soon. There's a lot to do, there's a lot to see, and the future is definitely going to look like the future. And I know that might seem a little bit intimidating, but here's something for you to remember as we end this video. You, my friend, watching right now, are already more up to date than the general population, which gives you somewhat of an advantage. And I know the sympathetic among you might say, hey, that's not really a great thing. We want to make sure everyone's caught up. And I'm with you, I'm trying. I'm making videos about it. But the truth of the matter is that a lot of people don't really understand AI. They don't know what it's capable of, and they don't know what's coming. So if you love your family, if you love your friends, educate them and continue to be optimistic about this technology. Because remember, there really is no reason to be negative. Sure, there's reason to be openly critical about things, but to think yourself into negativity about the future serves no purpose. I know, easier said than done. But trust me when I say I've done a lot of thinking about this stuff, and optimism really seems to be the best path forward for maintaining your mental sanity as the world gets crazier and crazier. So yeah, let me know what you guys think in the comments. I really want to have a great discussion with this. Also, please join my Discord server. There's a lot of AI focused folks on there who are thinking about this stuff as well and would love to talk about it. The more we discuss, the more we think, the more novel connections and the more prepared we can be for the upcoming explosion of AI technology. Thanks so much for watching. I'll see you in the next video. And goodbye.