# BREAKING: OpenAI INSIDER Drops BOMBSHELL "AGI Achieved" ## Метаданные - **Канал:** TheAIGRID - **YouTube:** https://www.youtube.com/watch?v=f9nveUUiRto - **Дата:** 23.11.2023 - **Длительность:** 20:13 - **Просмотры:** 153,108 ## Описание BREAKING: OpenAI INSIDER Drops BOMBSHELL "AGI Achieved" vWelcome to our channel where we bring you the latest breakthroughs in AI. From deep learning to robotics, we cover it all. Our 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 our latest videos. Was there anything we missed? (For Business Enquiries) contact@theaigrid.com #LLM #Largelanguagemodel #chatgpt #AI #ArtificialIntelligence #MachineLearning #DeepLearning #NeuralNetworks #Robotics #DataScience #IntelligentSystems #Automation #TechInnovation ## Содержание ### [0:00](https://www.youtube.com/watch?v=f9nveUUiRto) Segment 1 (00:00 - 05:00) so this might be one of the most important videos that we do make because this is concerning a topic that has been recently discussed in the artificial intelligence space and this is referring to a major breakthrough at openai from many Anonymous sources that are telling us that open ey have achieved the major breakthrough that many were speculating around last week which additionally led to the firing of the openi CEO Sam Altman and a whole host of many other decisions so right here you can see on this article it says open AI made an AI breakthrough before Sam Alman firing stoking a pyment and concern one day before he was fired by opening eyes board last week Sam mman alluded to a recent technical Advance the company had made that allowed it to push the veil of ignorance back and the frontier of Discovery forward I will be showing you guys that clip later because it is a real clip from Sam Alman that was recently made by himself so this isn't just some random person saying this these are comments from the CEO himself it says the cryptic remarks at the Apex CEO Summit went largely unnoticed as the company descended into turmoil but some open AI employees believe that alman's comments were referred to an innovation by the company's researchers earlier this year that would allow them to develop far more powerful artificial intelligence models a person familiar with the matter said the technological Breakthrough spearheaded by open AI Chief scientist Elia satk raised concerns among some staff that the company didn't have proper safeguards in place to commercialize such Advanced AI models and this is an article from the information the same place that was giving us all of the breaking news when open AI was going through quite the tumultuous period now we do know that this article does hold some Credence because not only did Sam Alman make numerous statements there have been numerous things that we discussed in previous videos that allude ude to the fact that AGI potentially and most likely has been achieved on some degree internally so let's take a look here one thing that we did want to include was the fact that Elon Musk also did take a look at this article and stated that it was extremely concerning now of course this makes sense because Elon Musk is an avid player in the AI space whilst owning companies like Tesla and of course his new company x. which recently launched the grock large language model which is said to compete with ch GPT but now let's take a look at the article that Elon Musk tweeted himself you can see that this is another article covering the exact same topic but from a little bit of a different angle it says open AI researchers wonn the board of AI breakthroughs ahead of the CEO aler sources say and this is of course from writers so this is on November the 23rd which is of course today and this was only a couple of hours ago now it says ahead of opening eyes see o Sam alman's aler several researchers wrote a letter to the board directors warning of a powerful AI discovery that they said could threaten Humanity to people familiar with the matter told Reuters now I find it interesting choice of words here that they said could threaten Humanity as that is something that is only largely discussed or talked about when people are saying that those who are discussing it are largely iders but if this is true then this could Mark the very next category and Evolution Ag large language models because this means that we've discovered something that would lead them to Greater capabilities so it continues to state that the previous unreported letter and AI algorithm were key developments before the board's aler of Alman the poster child of generative AI the two sources said prior to his triumphant return late Tuesday more than 700 employees had threatened to quit and join back Microsoft which is of course something that you do know now of course what was also interesting it says that the internal message to staffers referred to a project called qar or Q asterisk I'm just going to call it Q for now and a letter to the board before the events weekends one of the people said an OPI spokesperson said that the message sent by longtime executive Mira morati alerted staff to certain media stories without commenting on their accuracy so essentially what we have here is the name of this project which're just going to call it qar or Q asterisk so this is why you've been seeing on Twitter recently everyone has been talking about this because this is the code name now we don't know what it refers to but there are some interesting theories with as to what it true refers to and we're going to get into that later because there's a lot to discuss regarding the capabilities of this model and how exactly it works and why this is probably true the article continues to state that some at open AI believe qar could be a breakthrough in the startup search for what's known as artificial general intelligence AKA AGI one of the people told Reuters open AI defines AGI ### [0:00](https://www.youtube.com/watch?v=f9nveUUiRto) Segment 1 (00:00 - 05:00) so this might be one of the most important videos that we do make because this is concerning a topic that has been recently discussed in the artificial intelligence space and this is referring to a major breakthrough at openai from many Anonymous sources that are telling us that open ey have achieved the major breakthrough that many were speculating around last week which additionally led to the firing of the openi CEO Sam Altman and a whole host of many other decisions so right here you can see on this article it says open AI made an AI breakthrough before Sam Alman firing stoking a pyment and concern one day before he was fired by opening eyes board last week Sam mman alluded to a recent technical Advance the company had made that allowed it to push the veil of ignorance back and the frontier of Discovery forward I will be showing you guys that clip later because it is a real clip from Sam Alman that was recently made by himself so this isn't just some random person saying this these are comments from the CEO himself it says the cryptic remarks at the Apex CEO Summit went largely unnoticed as the company descended into turmoil but some open AI employees believe that alman's comments were referred to an innovation by the company's researchers earlier this year that would allow them to develop far more powerful artificial intelligence models a person familiar with the matter said the technological Breakthrough spearheaded by open AI Chief scientist Elia satk raised concerns among some staff that the company didn't have proper safeguards in place to commercialize such Advanced AI models and this is an article from the information the same place that was giving us all of the breaking news when open AI was going through quite the tumultuous period now we do know that this article does hold some Credence because not only did Sam Alman make numerous statements there have been numerous things that we discussed in previous videos that allude ude to the fact that AGI potentially and most likely has been achieved on some degree internally so let's take a look here one thing that we did want to include was the fact that Elon Musk also did take a look at this article and stated that it was extremely concerning now of course this makes sense because Elon Musk is an avid player in the AI space whilst owning companies like Tesla and of course his new company x. which recently launched the grock large language model which is said to compete with ch GPT but now let's take a look at the article that Elon Musk tweeted himself you can see that this is another article covering the exact same topic but from a little bit of a different angle it says open AI researchers wonn the board of AI breakthroughs ahead of the CEO aler sources say and this is of course from writers so this is on November the 23rd which is of course today and this was only a couple of hours ago now it says ahead of opening eyes see o Sam alman's aler several researchers wrote a letter to the board directors warning of a powerful AI discovery that they said could threaten Humanity to people familiar with the matter told Reuters now I find it interesting choice of words here that they said could threaten Humanity as that is something that is only largely discussed or talked about when people are saying that those who are discussing it are largely iders but if this is true then this could Mark the very next category and Evolution Ag large language models because this means that we've discovered something that would lead them to Greater capabilities so it continues to state that the previous unreported letter and AI algorithm were key developments before the board's aler of Alman the poster child of generative AI the two sources said prior to his triumphant return late Tuesday more than 700 employees had threatened to quit and join back Microsoft which is of course something that you do know now of course what was also interesting it says that the internal message to staffers referred to a project called qar or Q asterisk I'm just going to call it Q for now and a letter to the board before the events weekends one of the people said an OPI spokesperson said that the message sent by longtime executive Mira morati alerted staff to certain media stories without commenting on their accuracy so essentially what we have here is the name of this project which're just going to call it qar or Q asterisk so this is why you've been seeing on Twitter recently everyone has been talking about this because this is the code name now we don't know what it refers to but there are some interesting theories with as to what it true refers to and we're going to get into that later because there's a lot to discuss regarding the capabilities of this model and how exactly it works and why this is probably true the article continues to state that some at open AI believe qar could be a breakthrough in the startup search for what's known as artificial general intelligence AKA AGI one of the people told Reuters open AI defines AGI ### [5:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=300s) Segment 2 (05:00 - 10:00) as autonomous systems that surpass humans in most economically valuable tasks given the vast Computing resources the new model was able to solve certain mathematical problems the person said on condition of anonymity because the individual was not authorized to speak on behalf of the company though only performing math on the level of grade school students acing such chest made the researchers very optimistic about Q star's future success so we you do need to talk about how why acing such chest made researchers very optimistic about Q's future success one thing I've seen online and from many people on Twitter they're saying that Asing such test doesn't mean anything because if they can perform maths at the level of grade students this doesn't absolutely mean anything now I'd like to firmly disagree because the next slide is going to show you and this was pointed out by a tweet I don't remember which tweet this was but I want you all to see this because this is from opening eyes actual page so take a look at this piece of information so what we have here is something from open Ai and what you're looking at is something called predictability scaling so it says here a large focus of the GPT 4 project has been building a deep learning stack that scales predictably the primary reason is that for very large training runs like GPT 4 it's not feasible to do extensive model specific tuning we developed infrastructure and optimization that are very predictable Behavior across multiple scales to verify this scalability we accurately predicted in advance GPT 4's final loss on our internal code base which is not part of the training set by extrapolating from Models trained using the same methodology but 10,000 times less compute so you can see right here that opening eyes codebase next work prediction as it continues and continues we have the observed and then we have the prediction and of course we have GPT 4's abilities which means currently that essentially they were able to predict GPT 4 capabilities before training it and if you don't believe me take a look right here you can see that this is tweeted by Peter welinder he used to be the research leader openai and now he's the VP of product at open this is the guy that worked at open AI for 6 years he says important detail in gp4 blog post our research team were able to predict how smart it would be before it was trained and it is very likely that is what they are saying about qar or this new Secret model and this does make sense because not only did we get this blog post which is from GPT 4's blog post and Not only would get this article we also got some of Sam artman's statements so one of the statements that Sam Alman made you should check out which I will play now is that Sam Alman did talk about how they managed to discover some ability but they push the veil of ignorance back and they Advance forward for Humanity that's probably not the like for four times now in the history of open ey the most recent time was just in the last couple of weeks I've gotten to be in the room um when we sort of like push the front the sort of the veil of ignorance back and the frontier of Discovery forward and getting to do that is like the professional honor so now that you've seen that clip and that clip was very recent which is quite interesting considering all the information going around and I think it's rather fascinating that this statement by Sal Alman would be made at the same time that is these certain leaks and the other things are going on which I will continue to discuss now there was one statement that people are saying where Sam Alman said is this a tool or have we built a creature but it has been I think this is like definitely the biggest update year for people yet and maybe the biggest one we'll have because from here on like now people accept that powerful AI is going to happen and there will be incremental updates but there's like you know there was like the year the first iPhone came out and then there was like everyone since and at this point like you really know the difference between this one and last year's one um so this was like I think a big moment one thing that I am happy about is uh I think people are viewing these systems as correctly as tools artists in particular but um other people as well and there was this I think there's a real moment of fear which is like is this a tool we have built or a creature we have bu what is that going to mean I think people now view this as a new thing in the toolbox of humanity and are doing like really remarkable things with it it's very and I do want to firmly say that Sam Alman has previously multiple times stated that it's not a creature so any of these statements that people are saying I definitely think that this comment was taken out of context and I'm going to show you guys the full clip so you can see the entire context and not be confused but that doesn't mean that Sam Alman is isn't hinting or at least ### [5:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=300s) Segment 2 (05:00 - 10:00) as autonomous systems that surpass humans in most economically valuable tasks given the vast Computing resources the new model was able to solve certain mathematical problems the person said on condition of anonymity because the individual was not authorized to speak on behalf of the company though only performing math on the level of grade school students acing such chest made the researchers very optimistic about Q star's future success so we you do need to talk about how why acing such chest made researchers very optimistic about Q's future success one thing I've seen online and from many people on Twitter they're saying that Asing such test doesn't mean anything because if they can perform maths at the level of grade students this doesn't absolutely mean anything now I'd like to firmly disagree because the next slide is going to show you and this was pointed out by a tweet I don't remember which tweet this was but I want you all to see this because this is from opening eyes actual page so take a look at this piece of information so what we have here is something from open Ai and what you're looking at is something called predictability scaling so it says here a large focus of the GPT 4 project has been building a deep learning stack that scales predictably the primary reason is that for very large training runs like GPT 4 it's not feasible to do extensive model specific tuning we developed infrastructure and optimization that are very predictable Behavior across multiple scales to verify this scalability we accurately predicted in advance GPT 4's final loss on our internal code base which is not part of the training set by extrapolating from Models trained using the same methodology but 10,000 times less compute so you can see right here that opening eyes codebase next work prediction as it continues and continues we have the observed and then we have the prediction and of course we have GPT 4's abilities which means currently that essentially they were able to predict GPT 4 capabilities before training it and if you don't believe me take a look right here you can see that this is tweeted by Peter welinder he used to be the research leader openai and now he's the VP of product at open this is the guy that worked at open AI for 6 years he says important detail in gp4 blog post our research team were able to predict how smart it would be before it was trained and it is very likely that is what they are saying about qar or this new Secret model and this does make sense because not only did we get this blog post which is from GPT 4's blog post and Not only would get this article we also got some of Sam artman's statements so one of the statements that Sam Alman made you should check out which I will play now is that Sam Alman did talk about how they managed to discover some ability but they push the veil of ignorance back and they Advance forward for Humanity that's probably not the like for four times now in the history of open ey the most recent time was just in the last couple of weeks I've gotten to be in the room um when we sort of like push the front the sort of the veil of ignorance back and the frontier of Discovery forward and getting to do that is like the professional honor so now that you've seen that clip and that clip was very recent which is quite interesting considering all the information going around and I think it's rather fascinating that this statement by Sal Alman would be made at the same time that is these certain leaks and the other things are going on which I will continue to discuss now there was one statement that people are saying where Sam Alman said is this a tool or have we built a creature but it has been I think this is like definitely the biggest update year for people yet and maybe the biggest one we'll have because from here on like now people accept that powerful AI is going to happen and there will be incremental updates but there's like you know there was like the year the first iPhone came out and then there was like everyone since and at this point like you really know the difference between this one and last year's one um so this was like I think a big moment one thing that I am happy about is uh I think people are viewing these systems as correctly as tools artists in particular but um other people as well and there was this I think there's a real moment of fear which is like is this a tool we have built or a creature we have bu what is that going to mean I think people now view this as a new thing in the toolbox of humanity and are doing like really remarkable things with it it's very and I do want to firmly say that Sam Alman has previously multiple times stated that it's not a creature so any of these statements that people are saying I definitely think that this comment was taken out of context and I'm going to show you guys the full clip so you can see the entire context and not be confused but that doesn't mean that Sam Alman is isn't hinting or at least ### [10:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=600s) Segment 3 (10:00 - 15:00) giving us certain hints towards this AGI because we've known that Sam mman has made comments and made previous statements and then retracted them when they were also looked at further so one of the things that he also looked at after looking at these clips you can see that it says chat GPT boss says he's created human level AI then he said he's just Ming AGI has been achieved internally at openi Sam Alman writes on Reddit before backtracking and I did a whole video on this around two weeks ago where I took a deep dive into every single AGI claim and all these secrecies that were taking place many people hadn't realized that open ey secretly made several changes to their website to Showcase that they were going to be moving towards an AGI future and if you thought that was not weird you should take a look at some of the clips from that video which I will include now because they have a more Deep dive into some of these AGI specifics so you can see right here that currently the core values have shifted to AGI focus and it says we are committed to building safe beneficial AGI that will have a massive positive impact on Humanity's future anything that doesn't help with that is out of scope so that is crazy because they've put this as the first point meaning that currently something has changed to open AI to where they've decided that AI is going to be their North Star and by North Star I mean their focal point for where the company is going to go which means that systems like chat GPT and darly 3 are going to potentially be subparts of this AGI and what's interesting is that they said anything that doesn't help with that is out of scope so maybe this means that open ey was working on certain things and certain products that weren't necessarily going towards AGI but may still be AI based and now they've decided you know what we're not going to be working on that anymore we've now decided to just be focusing on AGI so that is why this is a Secret Bombshell that they secretly dropped silently updating their page and we now know that this is pretty insane now further on the video you might be thinking who on Earth is Jimmy apples and why does his statement on AGI even matter Jimmy apples if you didn't know is essentially an opening ey leaker we aren't sure of what he is or what he does but we do know that he always does have early information about GPT 4 or whatever open ey is currently working on if you're wondering about any skepticism regarding his statements previous statements around open ey have come out with 100% accuracy for example he actually tweeted and got the release date of GPT 4 a week before it was even announced meaning that he definitely has some kind of inside information now many sources speculate that this Jimmy Apple's guy is someone that spies on people and near the opening ey headquarters now you can see that there's a 10-minute video which goes into this and they talk about how Jimmy Apple says that let's pick a random date and March 14th and then of course GPT 4 is of course announced on that exact date we know that this person whoever they are definitely has some inside information now back to the statement about AGI he said that AGI has been achieved internally on September the 18th 2023 now what's crazy is that a week later the CEO Sam Altman tweeted that a has been achieved internally okay and he tweeted this on a certain post and then what was crazy was that he actually edited this comment and said obviously this is just meing when AGI is achieved it will be announced and it won't be announced with a Reddit comment so at the same time it seems that Sam Alman is aware of this opening eye leaker and he understands that whatever is going on behind the scenes that Jimmy apples is likely to know a decent amount about it so at the same time I find it super interesting that Jimmy apples goes ahead and says AGI has been achieved internally then a little bit over a month later open AI start to talk about AGI and state that AGI is simply their main focus and their core values anything that doesn't help with that is out of scope which leads me to believe that Sam Alman and his team at open AI are much closer to AGI than we initially think so we do have this tweet that breaks it down and I'm not going to read all of this because I don't want to confuse many of the standard viewers and sometimes you can get confused so I'm going to just try and break this down in the most simplest way possible but you can see right here that this is quoting the article and he says you need this type of thing to solve really hard math problems and it's really probably closer to reinforcement learning with AI feedback then we have a comment from Elon Musk that says it's some kind of alpha zero selfplay applied to large language models now if you don't know what he's referring to there he's referring to the fact that Alpha go which is one of the state-of-the-art models that was able to do crazy things including beating the alpha go world champion with unorthodox moves and doing ### [10:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=600s) Segment 3 (10:00 - 15:00) giving us certain hints towards this AGI because we've known that Sam mman has made comments and made previous statements and then retracted them when they were also looked at further so one of the things that he also looked at after looking at these clips you can see that it says chat GPT boss says he's created human level AI then he said he's just Ming AGI has been achieved internally at openi Sam Alman writes on Reddit before backtracking and I did a whole video on this around two weeks ago where I took a deep dive into every single AGI claim and all these secrecies that were taking place many people hadn't realized that open ey secretly made several changes to their website to Showcase that they were going to be moving towards an AGI future and if you thought that was not weird you should take a look at some of the clips from that video which I will include now because they have a more Deep dive into some of these AGI specifics so you can see right here that currently the core values have shifted to AGI focus and it says we are committed to building safe beneficial AGI that will have a massive positive impact on Humanity's future anything that doesn't help with that is out of scope so that is crazy because they've put this as the first point meaning that currently something has changed to open AI to where they've decided that AI is going to be their North Star and by North Star I mean their focal point for where the company is going to go which means that systems like chat GPT and darly 3 are going to potentially be subparts of this AGI and what's interesting is that they said anything that doesn't help with that is out of scope so maybe this means that open ey was working on certain things and certain products that weren't necessarily going towards AGI but may still be AI based and now they've decided you know what we're not going to be working on that anymore we've now decided to just be focusing on AGI so that is why this is a Secret Bombshell that they secretly dropped silently updating their page and we now know that this is pretty insane now further on the video you might be thinking who on Earth is Jimmy apples and why does his statement on AGI even matter Jimmy apples if you didn't know is essentially an opening ey leaker we aren't sure of what he is or what he does but we do know that he always does have early information about GPT 4 or whatever open ey is currently working on if you're wondering about any skepticism regarding his statements previous statements around open ey have come out with 100% accuracy for example he actually tweeted and got the release date of GPT 4 a week before it was even announced meaning that he definitely has some kind of inside information now many sources speculate that this Jimmy Apple's guy is someone that spies on people and near the opening ey headquarters now you can see that there's a 10-minute video which goes into this and they talk about how Jimmy Apple says that let's pick a random date and March 14th and then of course GPT 4 is of course announced on that exact date we know that this person whoever they are definitely has some inside information now back to the statement about AGI he said that AGI has been achieved internally on September the 18th 2023 now what's crazy is that a week later the CEO Sam Altman tweeted that a has been achieved internally okay and he tweeted this on a certain post and then what was crazy was that he actually edited this comment and said obviously this is just meing when AGI is achieved it will be announced and it won't be announced with a Reddit comment so at the same time it seems that Sam Alman is aware of this opening eye leaker and he understands that whatever is going on behind the scenes that Jimmy apples is likely to know a decent amount about it so at the same time I find it super interesting that Jimmy apples goes ahead and says AGI has been achieved internally then a little bit over a month later open AI start to talk about AGI and state that AGI is simply their main focus and their core values anything that doesn't help with that is out of scope which leads me to believe that Sam Alman and his team at open AI are much closer to AGI than we initially think so we do have this tweet that breaks it down and I'm not going to read all of this because I don't want to confuse many of the standard viewers and sometimes you can get confused so I'm going to just try and break this down in the most simplest way possible but you can see right here that this is quoting the article and he says you need this type of thing to solve really hard math problems and it's really probably closer to reinforcement learning with AI feedback then we have a comment from Elon Musk that says it's some kind of alpha zero selfplay applied to large language models now if you don't know what he's referring to there he's referring to the fact that Alpha go which is one of the state-of-the-art models that was able to do crazy things including beating the alpha go world champion with unorthodox moves and doing ### [15:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=900s) Segment 4 (15:00 - 20:00) things that nobody could predict it was because it was able to use something called H Tre search where it was able to think in a radically different way and because of that it was able to make radically different moves and essentially Elon Musk here is stating that it's going to use this kind of technology so it's going to be like Alpha zero selfplay applied to large language models and essentially what that was as well was that Alpha go actually played against itself millions and millions of times and then became the best player in the world it didn't actually need to play humans anymore it just played itself and iterated upon that data then of course we have the Q learning explanation so Q learning is basically a type of reinforcement learning which teaches computers to learn by rewarding them for making good decisions and then penalizing them for making bad ones this is something that you've heard before but this one is a bit different because it allows you to make long-term decisions so if we get into the six stages of this we have the environment and the agent which is where you have the environment like a video game or a maze and the agent which is the a or computer program that needs to learn how to navigate this environment then of course we have the states and the actions the environment is made up of many different states like different positions or scenarios in the game and then there's many different actions that you can take in the game like moving left or right or jumping then we have the Q table which is the cheat sheet that tells the agent what action is best to take in each state at the table is filled with the guesses because the agent doesn't know the environment yet then of course we have the learning by doing you start to explore the environment and every time it takes an action in the state it gets feedback from the environment it rewards the positive points and penalizes the negative points then of course it updates the Q table and then the Q table essentially considers the current reward and the potential future rewards this way the agent doesn't just learn to maximize immediate rewards but to consider long-term consequences over time it gets more and more accurate and it becomes better at predicting Wick's actions will yield the highest in rewards and eventually it can navigate the environment effectively so it goes on to State here and this is from gbt 4 that Q learning is like playing a complex video game where over time you learn the best moves and strategies to get the highest score initially you don't know the best actions to take but as you play the more you learn you're going to get better and better at the game and that's exactly what q-learning is so here are some of the tweets from The Wider AI community that have been on Twitter that are rather interesting that I thought I'd share and it says what could opening eyes breakthrough qar be about it sounds like it's related to Q learning for example qstar denotes the optimal solution of the Bellman equation alternatively referring to a combination of the AAR algorithm and the Q learning one of the natural guess is that it's alphao style Monte Carlo tree search of the token trajectory which is previously of course what we discussed and it says previously papers like Alpha code showed that even very naive root Force sampling in an llm can get huge improvements in competitive programming the next logical step is to search the token Tree in a more principled way so it says indeed qar seems to be about solving math problems then of course we have a very interesting clip here which is from Shane leg the founder and chief AGI scientist at Google Deep Mind and he actually talks about how this method that they're planning on using is pretty much exactly what we need in that next stage of evolution so take a look at that clip now these Foundation models are World models of a kind and to do really creative um problem solving you need to start searching so if I think about something like alphago in the move 30 famous move 37 where did that come from all its data that it's seen of human games or something like that no it didn't it came from it identifying a move as being quite unlikely but you know possible and then via a process of search coming to understand that the that was actually a very good move so you need to you to get real creativity you need to search through spaces of possibilities and find these sort of hidden gems that's what creativity is I think current language models they don't really do that kind of a thing they really are mimicking the data they mimicking all the human Ingenuity and everything which they have seen from all those data that's coming from the internet that's originally derived from humans if you want a system that can go be truly beyond that and not just generalize in novel ways so it can you know these models can blend things they can do you know Harry Potter in the style of a Kanye West rap or something even though it's never happened they can blend things together but to do something that's truly creative that is not just a blending of existing things that requires searching through a space of possibilities and finding these hidden gems that that are sort of the hidden away in there somewhere and that requires search so I don't think we'll see systems that truly Step Beyond their training data until we have powerful search in the process and then of course lastly we do have the Google Deep Mind founder Demis sabis saying that using a method called treesearch to explore with a large language model is largely going to be the next evolution of llms which they were planning to use in Gemini which they haven't released yet so if they managed to beat them to this it would be a surprise since ### [15:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=900s) Segment 4 (15:00 - 20:00) things that nobody could predict it was because it was able to use something called H Tre search where it was able to think in a radically different way and because of that it was able to make radically different moves and essentially Elon Musk here is stating that it's going to use this kind of technology so it's going to be like Alpha zero selfplay applied to large language models and essentially what that was as well was that Alpha go actually played against itself millions and millions of times and then became the best player in the world it didn't actually need to play humans anymore it just played itself and iterated upon that data then of course we have the Q learning explanation so Q learning is basically a type of reinforcement learning which teaches computers to learn by rewarding them for making good decisions and then penalizing them for making bad ones this is something that you've heard before but this one is a bit different because it allows you to make long-term decisions so if we get into the six stages of this we have the environment and the agent which is where you have the environment like a video game or a maze and the agent which is the a or computer program that needs to learn how to navigate this environment then of course we have the states and the actions the environment is made up of many different states like different positions or scenarios in the game and then there's many different actions that you can take in the game like moving left or right or jumping then we have the Q table which is the cheat sheet that tells the agent what action is best to take in each state at the table is filled with the guesses because the agent doesn't know the environment yet then of course we have the learning by doing you start to explore the environment and every time it takes an action in the state it gets feedback from the environment it rewards the positive points and penalizes the negative points then of course it updates the Q table and then the Q table essentially considers the current reward and the potential future rewards this way the agent doesn't just learn to maximize immediate rewards but to consider long-term consequences over time it gets more and more accurate and it becomes better at predicting Wick's actions will yield the highest in rewards and eventually it can navigate the environment effectively so it goes on to State here and this is from gbt 4 that Q learning is like playing a complex video game where over time you learn the best moves and strategies to get the highest score initially you don't know the best actions to take but as you play the more you learn you're going to get better and better at the game and that's exactly what q-learning is so here are some of the tweets from The Wider AI community that have been on Twitter that are rather interesting that I thought I'd share and it says what could opening eyes breakthrough qar be about it sounds like it's related to Q learning for example qstar denotes the optimal solution of the Bellman equation alternatively referring to a combination of the AAR algorithm and the Q learning one of the natural guess is that it's alphao style Monte Carlo tree search of the token trajectory which is previously of course what we discussed and it says previously papers like Alpha code showed that even very naive root Force sampling in an llm can get huge improvements in competitive programming the next logical step is to search the token Tree in a more principled way so it says indeed qar seems to be about solving math problems then of course we have a very interesting clip here which is from Shane leg the founder and chief AGI scientist at Google Deep Mind and he actually talks about how this method that they're planning on using is pretty much exactly what we need in that next stage of evolution so take a look at that clip now these Foundation models are World models of a kind and to do really creative um problem solving you need to start searching so if I think about something like alphago in the move 30 famous move 37 where did that come from all its data that it's seen of human games or something like that no it didn't it came from it identifying a move as being quite unlikely but you know possible and then via a process of search coming to understand that the that was actually a very good move so you need to you to get real creativity you need to search through spaces of possibilities and find these sort of hidden gems that's what creativity is I think current language models they don't really do that kind of a thing they really are mimicking the data they mimicking all the human Ingenuity and everything which they have seen from all those data that's coming from the internet that's originally derived from humans if you want a system that can go be truly beyond that and not just generalize in novel ways so it can you know these models can blend things they can do you know Harry Potter in the style of a Kanye West rap or something even though it's never happened they can blend things together but to do something that's truly creative that is not just a blending of existing things that requires searching through a space of possibilities and finding these hidden gems that that are sort of the hidden away in there somewhere and that requires search so I don't think we'll see systems that truly Step Beyond their training data until we have powerful search in the process and then of course lastly we do have the Google Deep Mind founder Demis sabis saying that using a method called treesearch to explore with a large language model is largely going to be the next evolution of llms which they were planning to use in Gemini which they haven't released yet so if they managed to beat them to this it would be a surprise since ### [20:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=1200s) Segment 5 (20:00 - 20:00) they've been working on this for quite some time with that being said let me know your thoughts on open ey's recent breakthrough has AI been achieved I personally do think so with everything said and if it has been achieved what do you think they're going to do next ### [20:00](https://www.youtube.com/watch?v=f9nveUUiRto&t=1200s) Segment 5 (20:00 - 20:00) they've been working on this for quite some time with that being said let me know your thoughts on open ey's recent breakthrough has AI been achieved I personally do think so with everything said and if it has been achieved what do you think they're going to do next --- *Источник: https://ekstraktznaniy.ru/video/14671*