Former Open AI Employee Reveals The Next 5 Years Of AI

So a former OpenAI employee has made some predictions in 2021 that have been scarily accurate up until 2024, 2025. He also recently made some predictions about the next five years of AI that I think you need to pay attention to. But before I actually dive into his predictions, I want to state that firstly, we're going to be taking a look at some of the predictions he made in 2021, because they actually describe what's happening right now with an eerie sense of accuracy. So let's take a look at some of Daniel Cocotaldjo's predictions. One of the first predictions that he made in 2021 was he said that in 2023 there's going to be insane hype. He said people are going to be continuing to talk about how these things have common sense understanding. Or do they? And there's also going to be lots of bitter think pieces arguing the opposite and how AI assistants and companions are are just around the corner. It's like self driving cars and drone delivery all over again. You can see right here making a prediction like this stating that, you know, the multimodal transformers are now even bigger. The biggest are about half a trillion parameters, costing hundreds of millions of dollars to train and a whole year and sucking up a significant fraction of the chip output of Nvidia. This is a remarkable prediction to make in 2021 with such startling accuracy. Remember, predicting the future is actually hardest in the AI industry because this is something that does grow exponentially and it's very hard for humans to visualize exponential. Now. He also made Another prediction about 2024, the year that we're currently living in right now. And looking at this prediction, it's fair to say that this is incredible. So it's clear here that he says the AI risk community has shorter timelines now, with almost half thinking some sort of point of no return will probably happen by 2030. This is partly due to various arguments percolating around, and partly due to these mega Transformers and the uncanny experience of conversing with their chatbot version. And what's crazy is that making this prediction in 2021, three years later, in 2024, many people are actually all stating that yes, by 2029, 2030, or by the end of this decade, there is going to be the singularity. That is a really interesting bet to make in 2021. Of course it's not a bet, it's a speculative blog post. But what I do want to state is that it's remarkably accurate and that's why? When we look into the future on the predictions that he's made about the coming years, 2026, 2027, 2028, we definitely shouldn't just think that they're that crazy, but that there might be some truth to them. Of course, he also says that the community begins a big project to build an AI system that can automate interpretability work. It seems maybe doable and very useful since pouring over neuron visualization is boring and takes a lot of personality. Now, what's crazy is that this is exactly what OpenAI are working on. They're actually working on how they can actually automate this interpretability research. And basically all that is, is that is just research that allows you to look inside of what an AI is actually doing and, and understand the decisions it's making. Now, let's take a look at some of the predictions that he recently made. Okay, so let's take a look at this. And this was a couple of months ago, so this is a screenshot. I'll leave a link down in the description below. But one of the predictions he's making recently about is 2024. So he says GPT Next release of GPT Next on Autonomous Agent. Likely to be available by the end of 2024. This model is expected to be a significant improvement over previous versions with enhanced capabilities and task completion and decision making. Now, I personally don't think that we're going to be getting agents this year. Of course, there is some nuance to this. It could happen a year later. Either way, a year later is still pretty fast. But what's actually interesting about this is that GPT Next is actually a real thing. If we actually take a look here at this graph released from an OpenAI secret presentation, we can see that there are three stages here. And of course, the final stage being something absolutely crazy. We've got the GPT3 era, which we had. Then, of course, we've got the GPT4 era, which we're currently in. And of course you can see just before we're about to go crazy into this GPT Next era. Now, this is probably going to be at the later stage of 2024, which is why I keep telling you guys, just wait until November, December, there might be a giant model release such as GPT Next that showcases what these models are truly able to do. Now, of course, it's not just OpenAI that's going to do this. Remember, they're not the only company that operates in the AI space. We've got companies like Google and Meta that are still playing in that space and can release models unexpectedly that surpass previous. What's crazy about this is that I don't think that this is going to be an autonomous agent. Although I could be wrong. AI breakthroughs can happen all the time that can literally speed up development tenfold. But the thing is, is that from what I've seen in interviews and discussions surrounding development of AI agents, reliability is still a factor and scale is still a factor. For example, if you're trying to get agents to do very well on certain tasks, the problem is that agents need reliable actions over many different tasks. Meaning that if you mess up just like 2% of the time, if you continue to perform actions where you mess up 2% of the time over the long term, the this actually means that you are very inaccurate. The point here is that reliability and scale are going to increase reliability in these models. We then take a look at 2025. This is where they speak about AI becoming widely adopted as personal assistants. These agents will be capable of performing various tasks, including making purchases. They will understand and execute complex instructions, significantly enhancing productivity and daily life management for users. Now, what's also fascinating about this is that we do know that that this is most likely penciled in. One of the things that I looked at when I was researching future models was I actually looked at the trademark Office for the GPT6 trademark. And in the GPT6 trademark, interestingly enough, what they actually have in that trademark is they have artificial intelligence agents. And basically that just essentially means that GPT6 is likely to be the product system that entails AI agents. And this makes sense because this also lines up with Daniel's prediction of autonomous agents being the year for 2025. Considering the fact that each iteration cycle probably takes 18 months, and considering GPT5 is nearing completion, the next cycle should be producing reliable AI agents which are going to completely transform certain parts of the economy. So I think that 2025 is most certainly going to be an interesting year because that is going to be where we potentially towards the later end of the year, have reliable AI agents that can perform tasks over long time horizons. I do think that it most likely might be Google who works on agents first, but I wouldn't be surprised if OpenAI get there too as there are some recent developments that I will talk about in new videos that are going to show you. These companies are a lot further ahead than you may think. Now, of course we have a look at 2026. In 2026, this is of course a crazy prediction but when you actually look at the rest of the predictions, it's not that crazy. And the reason I say that is because 2026, having a super intelligent AGI or the emergence of AGI that surpasses human level performance in most tasks is only two years away. Two years away from transformational technology seems like such a short time. Like I said before, humans have a bad perception of exponential growth. So you can see right here that it says this AGI will be capable of rapid learning and problem solving across diverse domains. And it's predicted that within 30 days of deployment, this AGI could reach a level to hundreds of humans experts. Now, I think this is rather fascinating because like I said before, one of the major predictions that I've looked at was the AGI prediction. One of the specific dates that I've continued to see from various sources. And this isn't just online speculations or online websites. These are actually research individuals, people working at top labs within OpenAI, Google, DeepMind and Anthropic. So the three Frontier labs, the main dates that I see are 2027 to the latest being 2030 for AGI. So 2026 is essentially just one year earlier. And it's not out of the question that this could potentially happen. Of course, there are many different things that could happen between then. There could be issues related to scaling, there could be some physical limits on what we're able to do. But in two years of development, towards the end of 2026, considering the fact that there is a lot more investment, a lot more competition now, it's not just the Western companies that are focusing on this. We've got China, that's there. It will be interesting to see what company manages to get to AGI first. Now we're going to go into 2027. So 2027 is where things start to easily get even crazier. So now 2027 does seem a bit crazy for artificial superintelligence. But if we do take a look back and consider the fact that if 2026 does actually get us to AGI, then getting to ASI wouldn't take that long. Because getting to ASI after AGI isn't that long, considering you're essentially automating AI research pretty much a hundredfold. Think of it like this. You currently we're moving at human speed, meaning that right now, in order to conduct AI research, there are a lot of things that we have to do. For example, a human wakes up, they have to get to work, maybe they drink their coffee, then they work all day, then they go home and they do other tasks. That's probably about six to eight hours of deep, focused work. But if you do have, for example, an AGI level system which is on par with a human, you could theoretically leave it running for 24 hours, meaning that you're likely to three times the output in a single day. But, but think of that. Over the course of a year, sometimes humans get ill, things happen, they're unable to work. There are things in the economy. But if we do get an autonomous system that is able to continuously advance AI research by itself, along with a few oversights from human intervention, I think it's going to be rather fascinating with how quick these developments could take place. And you have to understand that the main area where the compute is going to be focused on is, is of course, pouring into, just duplicating these AI systems. For example, you're not just going to have one smart AI system. That is something that most people don't realize. You're going to have millions and millions of copies of this artificial general intelligence that is working towards artificial superintelligence, which means that kind of exponential increase in terms of output towards AI research is going to be absolutely astounding for us to even comprehend. So you can see right here it says transition to ASI. So rapid advancement in AI capabilities potentially leading to an intelligence explosion. There's a 70% probability of ASI emerging by 2030. And this superintelligence is expected to solve global complex challenges and drive unprecedented technological progress. So this will be rather fascinating, a 70% chance of ASI emerging by 2030. And what's going to be interesting is how much those predictions change as we move towards that final date, towards the end of the decade. And whilst yes, this might not happen tomorrow, it might not even happen in 2026, but it is definitely a possibility. And when that does occur, which I do think it will, it's going to be a truly transformative time for the economy. Now, what comes after artificial superintelligence, because many people just think, okay, we're either going to be dead or we're going to be living in a technical utopia. But, but one of the things that most people don't realize because it's still in its infancy in terms of research and development, is of course, nanobots. So nanobots are microscopic robots that have a variety of different applications and use cases that could revolutionize many different industries. And it does say here that if ASI is not achieved, nanobots might emerge as transformative technology. There's a 30% chance of significant nanobot development by 2027 through to 2028. And these microscopic robots could revolutionize medicine, manufacturing and environmental remediation. Of course, nanobots can literally change environments, change humans. It's kind of strange how that sci fi kind of works. But once again, when we kind of take a look at these AI systems and if we were to grab them and show them to someone from 10 years ago, their mind would be blown. I mean, the first time I saw ChatGPT, I was definitely blown by what it was able to do. So this is of course crazy until it's not crazy. But of course, essentially the logic here is that if we have artificial superintelligence, it won't be hard for artificial superintelligence to develop ways and methods for nanobots to actually be commercially viable and for them to be economically viable in terms of actually working and changing the environment. And of course, if that does work, that is going to have remarkable implications for society. Now, of course, we do have a course 2029 being humanoid robots. Now, I think the reason that 2029 is the year for humanoid robots is because humanoids are facing the physical problem, okay? And that is because the physical world is a lot harder to master than the digital world. This is because collecting data in the physical world is time consuming. And right now we simply just don't have enough data to make these robots actually work well and be still effective. Now, of course, the other problem with humanoid robots is that they're just really expensive. Like some of the humanoid robots that you do see are upwards of $250,000. I mean, if you're going to make something like that available to the average person who wants one, or even to certain companies, they have to actually be commercially economically viable in the sense that they're going to be using that over a human. Unless that kind of robot is, you know, extraordinarily fast and is able to work for 24 hours on a single charge. People are not going to be spending $250,000 on a single robot. They're better off using the standard robots, the ones that are in factories, those single arms that are able to do repetitive tasks again and again. Or those other robots like the factories in Amazon where you can simply do picking and packing and just moving around boxes. Now, of course, if these humanoid robots are developed and if ASI is here, that kind of research is likely to speed up what we do in all areas. And one of the areas is going to be humanoid robots, which means that potentially we could be seeing embodiment or even better embodiment than we currently do have of current humanoids, which would bring us to a very sci fi level that many people currently do fear. Now I do not think that that is how humans go extinct. A robot runs off into the wild and just kills all of us. But I do think that this kind of embodiment is going to be there sometime in the future as robotics breakthroughs get there. Now what's also interesting is that Elon Musk actually did respond to this prediction and he said that Tesla will have genuinely useful human robots in low production in Tesla Internal next year and hopefully high production for other companies in 2026. So it is clear that there is a trend towards humanoid robots being increasingly part of the workforce and they actually do work pretty effectively if we take a look at what they're able to do in factories. But of course this is something that is very specific and it's very niche, so it's not something that can be applied to everything. Now with the Tesla bot, I do think it's pretty effective because if you've seen the demos, it's incredible at how effective it is to move. But of course there are a few limitations on what it can do with regards to mobility and many other factors. Now what I'm also going to show you guys here because I did actually make a video on this quite some time ago. I made a 30 minute video going over every single point from Daniel Coca Paljo in this prediction. But this one basically does say a few things that did have me quite surprised with what the predictions were because it shows us that if technology manages to continue to move at its current rate, we're going to see some incredible things. And one of the things that I never forgot was he says that whoever controls ASI will have access to a spread of powerful skills and abilities and, and we'll be able to build and wield technologies that seem like magic to us, just as modern technology would seem like magic to medieval. And this is something that's still hard for me to grasp, even as someone who understands that concept. Like, I know that right now if I grabbed like my phone and I went back to medieval times, that technology would seem like magic to them. Okay. And you know, when you think about it, a phone is kind of magic, you know, but of course I can't imagine there being technology that would seem like magic to me. It just feels, feels as if we're at the limit to where technology can be. But of course I know that this is not true. And this is just, you know, emotions or whatever, but thinking about that statement, the fact that they're going to have godlike powers over who doesn't control ASI is a rather fascinating statement because it implies that whoever gets the AGI first is probably going to have power over those who don't. And of course, at the top here you can see that probably whoever controls AGI will be able to use it to get to ASI shortly thereafter, maybe in another year, give or take a year. So it's pretty crazy on what's going on here. And I'm not going to lie, guys, there is a lot of stuff coming in the future that you should definitely be paying attention to because all of these technologies are going to impact you in one way or another. So if you enjoyed this video, let me know what you think your predictions are. Do you think his predictions are pessimistic? Do you think they are too optimistic? Let me know what you think about the predictions for the future. I would love to know if you think the future is going to be slower than we think or if it's going to be faster than we think. With that being said, if you enjoyed the video, don't forget to, like, comment down below and I'll see you all in the next one.