# STUNNING GPT-5 Details Are Finally Released....

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=4qdKfpBFals
- **Дата:** 11.07.2024
- **Длительность:** 22:03
- **Просмотры:** 76,573
- **Источник:** https://ekstraktznaniy.ru/video/14194

## Описание

Learn A.I With me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Links From Todays Video:
https://www.bloomberg.com/news/articles/2024-07-11/openai-sets-levels-to-track-progress-toward-superintelligent-ai
https://arxiv.org/pdf/2311.02462
https://www.perplexity.ai/search/what-is-a-doctorate-level-educ-3ebWYnewQJuzWLER1akzPA
https://x.com/tsarnick/status/1803901130130497952
https://x.com/Coulmock_Yov/status/1798984934142709996
https://x.com/search?q=TSARNICK%20AGENT&src=typed_query
https://x.com/tsarnick/status/1808252234494574634/video/1
https://www.bloomberg.com/news/articles/2024-03-22/openai-courts-hollywood-in-meetings-with-film-studios-directors
https://video.twimg.com/ext_tw_video/1808251994286837762/pu/vid/avc1/720x720/KwXyAnmo_nxL2ejb.mp4?tag=12
https://video.twimg.com/ext_tw_video/1803900886760226816/pu/vid/avc1/720x720/KRsZbT1ebDzMVLdN.mp4?tag=12
https://video.twimg.com/ext_tw_vi

## Транскрипт

### Segment 1 (00:00 - 05:00) []

so open AI have finally given us what they call the next level in terms of AI we finally got the details on what GPT 5 is going to be called SLG graded at it the reasoners that's what they're calling the next level of AI systems and they're basically saying that this is going to be PhD levels and there were actually a few secret things leaked so let's get into exactly what's going on we can see here that Bloomberg have recently reported only around an hour ago about this insane article to where open AI has developed a syst system to track progress towards human level AI the company believes its technology is approaching the second level of five on the path to artificial general intelligence now whilst that might not seem like some kind of monumental change I'll show you all why this is a lot more important than you may have thought so let's take a look and dive into this article to show you guys why this is pretty crazy so the article reports that open AI has come up with a set of five levels to track its progress towards building artificial intelligence software capable of outperforming humans the startup's latest effort to help people better understanding thinking about the safety and the future of AI the chat GPT maker seen by many as a leader in the race to build more powerful AI systems shared the new classification systems with employees on Tuesday during an all hands meeting an openi spokesperson said and the tears which openi plans to share with investors and others outside the company range from the kind of AI available today that can interact with in conversational language with people to AI that can do the work of an organization level five now this is where things start to get crazy because they're talking about going from level one which is what we have now which is stuff like GPT 4 GPT 3. 5 clae 3. 5 to all the way until level five which is the work of an organization now when you start to think about the implications of an AI doing the work of an organization you can start to wonder why this is such a big deal it goes on to read the openi told employees that the company believes it is currently on the first level according to the spokesperson but is on the cusp of reaching the second which it calls reasoners so that is going to be the new term that openi is using in order to discuss exactly what's going on this is what they call reasoners so they said that this refers to systems that can do basic problemsolving tasks as well as a human and this is probably the point that you should take away from this video and this is going to be something that I continue to emphasize is that it can do tasks as well as a human with a doctorate level education who doesn't have access to any tools now this is pretty crazy but why is this pretty crazy so the reason this is so incredible is because open AI are working on something that is basically at the level of a doctoral degree this is the highest level of academic degree awarded by universities and other educational in institutions this signifies a high level of expertise and a specific field of study or profession there are two primary types of doctoral degrees academic which are research focused and professional which are practice focused so you've got the academic doctorates which are the doctor of philosophy the PHD and this is the most common type of academic doctorate focusing on original research the creation of new knowledge in a specific field and it involves several years of coursework comprehensive exams and the completion of dissertation based on original research and of course you've got other academic doctorates and these include degrees like doctor of Education doctor of theology doctor of the Fine Arts which also emphasize research but may have specific focuses within their fields and when we take a look at the true definition of what a doctorate is an AI system being able to do this is truly remarkable because we didn't know that this kind of jump was going to be there and there were some initial things that people did say that were actually hinting towards this that you might have actually missed there are two main clips that I've seen that showcase just how crazy this is and now we know that for 100% certainty that the people at open Ai and some of the people at Microsoft weren't lying when they said that GPT 5 is likely to be at the PHD level in this clip you'll actually see Kevin Scott Microsoft's CTO regard GPT 5 or whatever model they're going to call it on the level of phds basically stating that what open AI just said is actually pretty true now this actually happened around a week ago not that long ago but it was met with early skepticism because people didn't really know whether or not to trust them because of course open AI have been criticized in the past for you know talking about certain products but not essentially releasing them we're seeing technically all of the things fall in place to have uh like really durable memories uh with the systems uh like reasoning I think gets a lot better so you know some of the fragility in the

### Segment 2 (05:00 - 10:00) [5:00]

current models are you know can't solve very complicated math problems and has to you know sort of bail out to other systems to you know do very complicated things I mean the way that maybe I would phrase it is if you think of gpte 4 and like that whole generation of models as things that can uh perform as well as a high school student on things like the AP exams um uh some of the early things that I'm seeing right now with the new models is like you know maybe this could be the thing that could pass your qualifying exams when you're a PhD student this is another clip of Mira morati open ai's CTO where she was basically stating that gpt3 was Todd level gp4 was a smart high schooler and in the next gen to be released in a year and a half will be PhD level now this once again was met with some early criticisms because many people were speculating that this was just the hype and of course since open AI haven't truly released any of their mind-blowing products just yet many people are starting to speculate whether or not openai is living up to what they claim they are and considering many other model releases have been you know recently released that are somewhat better than gp40 they're spectating they're speculating whether or not this is going to be true but now we know that when open a that the CTO or any members do say things we can now be like okay we don't want to take it with a grain of salt anymore because we know that they're pretty much spot on these systems are already human level in specific tasks um and of course in a lot of tasks they're not if you look at the trajectory of improvement um systems like gpt3 where maybe uh let's say you know toddler level intelligence and then systems like gbd4 are more like smart high schooler in Ence um and then in the next couple of years we're looking at PhD level intelligence for specific tasks um so things are changing and improving pretty rapidly MH meaning like a year from now yeah a year and a half let's say yeah clip not only gave us the PHD level of course which is pretty remarkable when you come to think about the economic impacts of that and of course we did get apparently a timeline which says a year and a half so that means that we could potentially getting this model in pretty much late 2025 but it will be interesting to see how that kind of system is deployed I do think when you do talk about PhD level intelligence I think this will be firstly restricted most likely to research organizations and the reason I do state this and I know I'm going off on a tangent here but just hear me out cuz what I'm about to say is probably going to happen but if we take a look at this article right here you can see that open AI caught Hollywood in meetings with film studios and directors now let me give you guys a quick rundown as to why this news is important if you remember Sora was the technology that was released and people were like oh my God this is crazy this is amazing however we never got Sora we haven't got it we haven't got access we don't know if we're ever going to get access now why is that important well if gp5 is PhD level what that could mean is that it's a very powerful system and if it's so powerful openi might think okay we can't just give this out to the public because they might just jailbreak it and then we're going to get into trouble so similarly to what they' done with Sora they might just go to some big corporate clients where they can actually use these models and actually profit heavily from them so basically with what they're doing with Sora is that open ey are courting Hollywood Studios in meetings with film directors and film studios and they're basically saying that look we can't releas it to the public because number one they're going to be mad at us and number two they might jailbreak the system but what we can do is we can give it to you guys can just use it to make your films and I guess it's a win-win for them now I think this might happen with GPT 5 because they might have a really like extreme internal model that is just so smart that it's only able to be released to you know research organizations or Charities that are doing some humanitarian work and whilst yes this is pretty crazy in terms of the kind of release that they could do I don't think it is out of the question you can also see here that you know open AI has a deliberate strategy of working in collaboration with industry through a process of iterative deployment rolling out AI advances in phases in order to ensure safe implementation and to give people an idea of what's on the horizon we look forward to an ongoing dialogue with artists and creatives now since they've done that with Sora I do expect them to be doing something very similar with these Advanced models as iterative deployment is exactly what openi is now about and the reason that this is so

### Segment 3 (10:00 - 15:00) [10:00]

crazy is because you can see that here it says that this is like someone who doesn't have any tools so imagine being able to do PhD level work without having access to any tools I think that is something that you might miss but it's something that you really don't want to you know Miss because it's truly important because what this also potentially States is that not only is the model that much better at reasoning that potentially some of the agentic Frameworks that we previously saw in some of the previously you know demos and videos from people like Andrew and G that focus on agentic capabilities and pushing the boundaries of what's currently capable with the current systems I think maybe those are either now internal into the model or they've probably wrapped those agentic workflows natively into GPT 5 which pushes those capabilities even further so this is going to be truly impressive SL remarkable because we're about to see how crazy things get article continues to say that at the same meeting company leadership gave a demonstration of a research project involving its GPT 4 AI model that open AI thinks shows some new skills that rise to humanlike reasoning according to a person familiar with a discussion who asked not to be identified because they were not authorized to speak to press so essentially here once again we also got something which is wrapped around GPT 4 that apparently gets to human level reasoning so once again if you're unfamiliar with what this is basically there's ways that you can use GPT 4 in an agentic framework and that just basically means getting it to think in a certain way and getting it to do tasks in a specific way for example if you ask GPT 4 to think step by step before reading out its responses it usually improves the response quite a lot and what we can see here is that there has been an internal demonstration of a research project involving just GPT 4 that means that it rises to Human like reasoning According to some people so we don't know what this is or what this could be but whatever it is it shows us that open ey have once again managed to do something that clearly breaks the mold in terms of what is you know conventionally capable of in terms of these systems so that thing right there I do wonder what that is I don't know if it's qar or what not I do think if it was qar we would probably have just heard that it's qar but I do think that this is something important now of course the article continues to stay that opening ey for years has said that it's working to build what's often referred to as artificial general intelligence or AGI essentially computers that can do a better job on most tasks than most people and such systems do not currently exist and of course he expects that AGI could be reached within this decade that's basically Sam alman's predictions so here's where we get onto the actual meat of the video and there was a lot of meat in this article which is rather surprising they said here that AI researchers have long debated on how to deter determin the criteria for reaching AGI in a November 2023 paper which is one from Google deepmind several researchers at Google Deep Mind proposed a framework of five ascending levels of AI including tiers such as expert and superhuman the rankings resemble a system often referred to in the automotive industry in order to assess the degree of automation for self-driving cars so basically there was this paper from Google Deep Mind which I'm going to show you guys all now and this was basically the paper I'm just going to zoom in a bit here but it calls levels for AGI operationalizing progress on the path to AGI basically what they State here is that AGI is pretty confusing and they need to sort of map out how we can start to measure AGI because some people think gp4 is Sparks of AGI but of course it's not and you know some people would say that it's fully AGI which of course it's not some people say it's just baby AGI there's all these definitions and people would like a complete definition on where we are so here we have the table you can see that this is basically how they you know quantify SL rank AGI SL AI systems in terms of how capable they are and you can see that currently we do have five levels here but these five levels aren't the levels that I'm talking about slop AI talking about these are just basically similar to what open ey has recently done and you can see that you know there's level two which is competent which is at least 50% of skilled adults expert at least 90% virtuoso which is 99 % and of course superhuman which outperforms 100% of humans and we can see that you know currently we don't have competent AGI yet which is unfortunate but eventually we will get there and of course you can see that it's basically you know expert AGI is the one that's probably going to change the world and virtual also AGI is pretty crazy and artificial super intelligence is uh something that we don't even need to talk about yet so we can see here that what we are looking at are five different stages to artificial intelligence and the reason that this is important is because this is opening

### Segment 4 (15:00 - 20:00) [15:00]

eyes internal way of you know gauging how capable their systems are so we can see here that level ones are chatbots with conversational language and the level that's got everyone talking is of course level two which are reasoners with human level problem solving now I think this is a lot bigger than people think because getting human level problem solving from an AI system is something that is currently pretty hard to do because of the architecture of LMS these models are pretty narrative which means they predict the next token and in doing so they don't develop you know real world you know understanding of certain things and some people argue that these systems are just pretty much predicting the next word which means that they're just stochastic parrots now there's a lot of evidence to support this and against this however if we can manage to use these systems to get to reasoners which have human level problem solving either be a tree search or whatever agentic capabilities that you might want to have I do think that this is going to be a big deal because what this does is this increases the reliability of the model and this allows it to be applied to many more different Industries which means the economic impacts and the kind of use cases that this AI system has is going to be vastly more impactful so you can see here that some of the other tiers that we do have here are systems like agents systems that can take actions which are essentially autonomous systems that can spend you know days acting on a user's behalf as you can see right here and level four describes AI that can come up with new Innovations and the most advanced level would be organizations so I think when we actually take a look at what's going on here I think this shows us that even though this might be you know iteratively developed systems I think each level is going to be even more crazier than the last because going from you know nothing to a chatbot is pretty crazy in its own regard talking to an AI system all day is something that you can easily do with a remarkable level of accuracy and then going to a Reasoner that has human level problem solving is going to be massively impactful because solving problems is something that you know these current AI systems struggle to do which means that if this is human level problem solving at a PhD level that means it's likely to pretty much decimate all of the existing benchmarks that these current systems do struggle with and it's going to be pretty crazy to see what happens here because we haven't heard any of the other companies talk about their future systems in the way that open AI is and if we take a look at agents that's going to be autonomous systems that can take actions and recent reports show that of course this is I think about two years away and essentially going to be there in the next level and if you don't believe me I will show you guys a clip from that I'll add to the video if you want an agent to act in the world um usually that acting requires you to you know engage in a series of actions right you talk to a chatbot it only answers and maybe there's a little followup but with agents you might need to take a bunch of actions see what happens in the world or with a human and then take more actions and so you need to do a long sequence of things and for that to actually work the error rate on each of the individual things has to be pretty low right if I'm a robot and I'm like you know okay I'm going to pick up this thing and walk over there and I'm going to pick up that you know I'm building the house or something there's probably thousands of actions that go into that and so all of this is to say the models need to get more reliable because the individual steps need to have very low error rates and I think part of that will come from scale um like we need another generation or two of scale before the agents will really work but I think level four which are the innovators that you know AI that can Aid in invention is going to be absolutely incredible because an basically be an AI that could possibly Aid in AI research and if an AI can in AI research it means that AI research is going to speed up quite a bit and if that happens we know that feedback loop of AI aiding in AI research that makes better AIS that can do even more AI research that's going to be where things start to go very crazy and I think the last level organizations is one that is even crazier because organizations is something that I truly cannot even comprehend just yet if open AI are saying an AI that can do the work of an organization is one of their level five goals and whilst yes they essentially on the brink of level two I think yes level five is going to be some time away but the fact that is even a possibility shows us that what we're moving towards is a truly new future where things are just going to be absolutely incredible now you can see here that they said that the levels were put together by Executives and other senior leaders at openi and it is considered a work in progress and the company will gather feedback from people including employees investors and its board and make tweak the levels over time so let me know what you think about this where you think we are on this I

### Segment 5 (20:00 - 22:00) [20:00]

mean you know some of this stuff isn't going to be coming for too long but when I say too long remember that's just in the AI world and that means it's only a year or two before you know rapidly capable systems are deployed whether externally or internally I still think the level of development in these companies and of course whatever's going on here is going to be crazy now there's one last thing I do want to say because I think it might be true here one of the things that I do think is true here or that could be kind of interesting is the fact that they didn't actually state that this was built on top of GPT 4 whilst yes it probably is going to be built like a GPT 5 level model one of the interesting things that I saw here and that I'm picking up on is that they are calling this reasoners and this refers to systems that can do basic problem solving tasks as well as a human can with doctorate level education who doesn't have access to any tools so I'm speculating that maybe this model might not be like GPT 4 maybe the internal workings of the model are probably going to be extremely different and I do wonder how that model is going to be scaled and how it's going to be deployed because they didn't really talk about it as being you know just like GPT 5 level in terms of just you know another model that they've scaled up but I think that they're definitely exploring different ways to use this model in terms of How It's reasoning about its problems and one of the things that Sam Alman did previously talk about in an interview with Bill Gates was that you know if GPT 40 gives you an answer sometimes we don't know whether not that answer is good but if you could take a thousand of those answers and then pick the top three I guess that one would be a lot better in terms of the reliability so maybe we're going to see something like that in these future models that is being done internally the only problem though that I would say is that open eye doesn't release their research so even if we do get reasoners you know considering that they're not calling gb5 a chatbot anymore it seems that what we're going to get we're not truly going to understand the inner workings of it we're just going to see demos and the results of it