# Former Open AI Employee Reveals The Next 5 Years Of AI ## Метаданные - **Канал:** TheAIGRID - **YouTube:** https://www.youtube.com/watch?v=bTxm49rHi68 - **Дата:** 05.08.2024 - **Длительность:** 17:54 - **Просмотры:** 26,984 ## Описание Prepare for AGI with me - https://www.skool.com/postagiprepardness 🐤 Follow Me on Twitter https://twitter.com/TheAiGrid 🌐 Checkout My website - https://theaigrid.com/ 00:00 - Introduction 00:49 - 2021 predictions for 2023 (AI hype, large language models) 02:10 - 2021 predictions for 2024 (AI risk community timelines) 03:12 - Recent predictions for 2024 (GPT next, autonomous agents) 05:22 - Predictions for 2025 07:34 - Predictions for 2026 09:25 - Predictions for 2027 ( 11:41 - Discussion on (2027-2028) 13:19 - Predictions for 2029 15:33 - Elon Musk's response to humanoid robot predictions 16:14 - Implications of controlling Artificial Superintelligence 17:34 - Closing thoughts and request for viewer opinions Links From Todays Video: https://www.reddit.com/r/singularity/comments/1e9d5pb/former_openai_researcher_predictions/#lightbox Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos. Was there anything i missed? (For Business Enquiries) contact@theaigrid.com #LLM #Largelanguagemodel #chatgpt #AI #ArtificialIntelligence #MachineLearning #DeepLearning #NeuralNetworks #Robotics #DataScience ## Содержание ### [0:00](https://www.youtube.com/watch?v=bTxm49rHi68) Introduction so a former open AI 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 5 years of AI that I think you need to pay attention to but before I actually dive into his predictions I want to sayate 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 cokal prediction 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 ### [0:49](https://www.youtube.com/watch?v=bTxm49rHi68&t=49s) 2021 predictions for 2023 (AI hype, large language models) assistants and companions 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 and 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 2020 4 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 2013 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 3 years later in 2024 many people are actually all stating that yes by 2029 sl230 or by the end of this decade there is going to be the singularity ### [2:10](https://www.youtube.com/watch?v=bTxm49rHi68&t=130s) 2021 predictions for 2024 (AI risk community timelines) 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 should 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 person hour now what's crazy is that this is exactly what open AI are working on they're actually working on how they can actually automate This interpretability research and basically all that is that is just research that allows you to look inside of what an AI is actually doing and understand the decisions it's making so 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 ### [3:12](https://www.youtube.com/watch?v=bTxm49rHi68&t=192s) Recent predictions for 2024 (GPT next, autonomous agents) description below but one of the predictions he's making recently about is 2024 so he says GPT next released of GPT next an 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 in 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 open AI 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 GPT 4 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 open ey 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 turn fold but the thing is that from what I've seen in interviews and discussions surrounding development of AI agents reliability is still a factor and scale 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 this actually means that you are very inaccurate so the point here is that ### [5:22](https://www.youtube.com/watch?v=bTxm49rHi68&t=322s) Predictions for 2025 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 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 GPT 6 trade Mark and in the GPT 6 trademark interestingly enough what they actually have in that trademark is they have artificial intelligence agents and basically that just essentially means that GPT 6 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 GPT 5 is near Inc 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 longtime Horizons I do think that it most likely might be Google who works on agents first but I wouldn't be surprised if open AI 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 intelligence AGI or the emergence of AGI that surpasses human level performance in most tasks is only 2 years away from you know transformational technology seems like such a short time ### [7:34](https://www.youtube.com/watch?v=bTxm49rHi68&t=454s) Predictions for 2026 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 rele 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 AI 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 2 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 get even crazier so now 2027 does seem a bit crazy for artificial super intelligence but if we do take a look back and consider F 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 100 fold think of it like ### [9:25](https://www.youtube.com/watch?v=bTxm49rHi68&t=565s) Predictions for 2027 ( this currently we're moving at human speed meaning that right now in order to conduct 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 6 to 8 hours of deep SL 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 think of that over the course of a year sometimes humans get ill things happen they 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 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 real realize you're going to have millions and millions of copies of this artificial general intelligence that is working towards artificial super intelligence 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 rapid advancement in AI capabilities potentially leading to an intelligence explosion there's a 70% probability of ASI emerging by 2030 and this super intelligence is expected to solve Global complex challenges and drive unprecedented technological prog 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 super intelligence 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 one of the things ### [11:41](https://www.youtube.com/watch?v=bTxm49rHi68&t=701s) Discussion on (2027-2028) 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 achieve Nanobots might emerge as transformative technology there's a 30% chance of significant nanobot development by 2027 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 chat GPT 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 super intelligence it won't be hard for artificial super intelligence 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 of 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 ### [13:19](https://www.youtube.com/watch?v=bTxm49rHi68&t=799s) Predictions for 2029 this is because collecting data in the physical world is timec 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 SL 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 20 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 and 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 is how you know 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 ### [15:33](https://www.youtube.com/watch?v=bTxm49rHi68&t=933s) Elon Musk's response to humanoid robot predictions 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 kokalo 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 ### [16:14](https://www.youtube.com/watch?v=bTxm49rHi68&t=974s) Implications of controlling Artificial Superintelligence never forgot was he says that whoever controls ASI will have access to a spread of powerful skills and abilities and will 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 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 ### [17:34](https://www.youtube.com/watch?v=bTxm49rHi68&t=1054s) Closing thoughts and request for viewer opinions 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 that being said if you enjoyed the video don't forget to leave a like comment down below and I'll see you all in the next one --- *Источник: https://ekstraktznaniy.ru/video/14146*