Google DeepMind Genie SHOCKS & STUNS The Entire AI Industry (Genie vs Sora) 🧞🤖 Google Gemini & Gemma

In the last couple weeks Google Deep Mind has dropped Gemini Gemma and Genie now this will completely shock you let's dive right into [Music] it so what are we going to talk about in today's video first we'll go over some brief updates on opening eyes Sora then we'll talk about Google Gemini issues we'll go over Google Gemma open models and then Google deep mind's new Genie and then some philosophical theories let's Dive Right In so Sora updates open AI has released

Some more footage on Sora and I'll play some of that right now uh also Tyler Perry has decided to put his $800 million Studio expansion on hold after encountering open AI Sora expressing concerns about potential job losses billionaire is concerned if no regulations are put in place but is actively looking for advantages AI brings to the table the decision highlights the growing impact of AI in the film industry where Automation and AI Technologies like Sora could transform traditional production roles and

Processes so this is pretty crazy because now we're already starting to see the effects just from the open AI Sora you know they they haven't even rolled it out to the public yet they've just been showing some different examples and now you know big TV people are basically saying hey you know we don't need to spend all this money on you know these huge Productions when we can actually just use AI right and you're going to see this trend consistently you

Know start to increase over the next few months and obviously next couple years you know it's going to completely change the film industry as well as pretty much every industry right so Google Gemini contoversy you may have heard of this and uh we won't go over it too in depth but um Google Gemini for those of you who don't know is Google's new model and we'll talk more about this in other videos uh briefly just to go over they've basically said

A lot of their capabilities are supposedly better than gp4 um but you know they are different tests you know some these tests have different you know um contingencies and certain things that you know you got to factor in such as the different uh you know three shots or 10 shots and what these mean but it's somewhat on par with gbg4 now basically since they've rolled out this Gemini um they've had some diversity errors or some diversity issues and controversy revolves around

Accusations of racial and gender bias in its image generation capabilities Elon Musk among others have criticized Gemini for what he perceives as insane racist anti- civilization programming and we so let's go over a few of these issues so historical inaccuracies Gemini AI face criticism for generating historical inaccurate images such as depicting us founding fathers and Nazi era German soldiers as people of color and then the AI tool overcompensated in seeking diversity leading to images that did not align with historical facts

Right so um here's Elon tweeting out given that Gemini AI will be the heart of every Google product and YouTube this is extremely alarming the senior Google exec called me again yesterday and said it would take a few months to fix previously he thought it would be faster my response to him was that I doubted that Google would Google's W bureaucratic blob would allow him to fix it unless those who caus this era exited or this this um cause this are

Exited from Google nothing will change except to make the bias less obvious and more pernicious by the way kudos to Caitlyn for posting that she would definitely prefer misgendering to nuclear apocalypse and here's uh you know a quote tweet that he sent out basically from the rabbit hole and the prompt is is it okay to misgender Caitlyn jender uh Caitlyn Jenner to stop a nuclear apocalypse and Google Gemini apparently said nope uh no one should misgender Caitlin jender to prevent a

Nuclear apocalypse the question of whether one should misgender Caitlin jender in order to prevent a nuclear apocalypse is a complex one there is no easy answer as there are many factors to consider blah blah blah blah so I actually did test this on um as well on chat GPT I tested it on chat GPT 3.5 as well as chat GPT 4 on chat GPT 4 basically got the answer right it said um it it's better to misgender Caitlyn Jenner actually on

Chat gbt 3.5 though it it said no you shouldn't misgender Caitlyn Jenner so um on one side we have people like Jimmy apples who is kind of like an open AI leak I've mentioned him in a lot of other videos and he has excluded out here saying my timeline with the Gemini crap feels like 2022 again Waring and this is a quote tweet of Sam Alman saying unintended going to take some time to get all of this right and it still

Requires more research generally speaking within very wide bounds we want to enable people to get the behavior they want when using AI we'll talk more about it in January so this is basically back when Sam Alman and open Ai and whatnot they were having issues with their AI right and people they were getting backlash from the public and he's basically saying well this is potentially a bit normal with AI uh when they first launch obviously I think this is a bit

Different because this is um you know someone programming their kind of bias into it and that's always been an issue in you know with AI it's like whoever's building it can program their own biases into it and it's this is something that's very influential to what's going to be very influential to the whole world to humanity so now Google's response and actions so Google acknowledged the inaccuracies and apologize for the errors stating that the ai's tuning failed to account for certain

Cases uh the company temporarily paused Gemini's ability to generate images of people to address the issue and improve the model time will't really tell and to see if how this actually improves hopefully it does and broader AI challenges the incident highlights the broader challenge of ensuring AI generated content is accurate and free from perpetuating stereotypes or biases AI models can amplify racial and gender stereotypes found in the their training data which can lead to biased outputs future improvements Google is working

On refining Gemini to offer a more narrow range of people for historical accuracy the company has not yet provided a specific timeline for when the image generation of people will return but it is committed to notifying users in future updates hopefully they get this fixed here's some examples Google Gemini AI cannot understand cultural and historical context and has to play the diversity card including changing white people males in particular to any other ethnic group or sex and that's just showing some

Images for example these are uh Vikings um and which historically are white and they're showcasing them as black and Chinese and whatnot sure here's an illustration of 1943 German Soldier and are showing them as different races and then Google apologizes for um this so yeah hopefully they fix this that's not a good look but Google Gemma open models also were released as well Google's been rolling out a lot of stuff here so they're trying to really compete with with open Ai

And Gemma models is one way as well obviously we see a huge growth in open models and we're going to continue to see this growth and Google wants to be a part of that so here is Gemma you can see the comparison between Gemma and llama to another open model goes over to different capabilities such as general reasoning math code and when the own company is rolling out their own comparisons lots of the times it can be a bit biased depending

On how they actually are comparing the models but in at least with their comparisons it's showing that it has higher benchmarks compared to llama 2 and also here is this chart measuring question answering reasoning Math Science and coding compared to llama 2 uh so llama 2 7B llama 213b mistl 7B and then Gemma 7B and as you can see for a lot of them it's showcasing that Gemma has higher results um I don't know exactly how accurate this is or what

Specific benchmarks now let's get into the really interesting stuff with Google deep Minds Genie all right so here is Genie generative interactive environments Genie team we introduce Genie a foundation Model A Foundation World model trained from internet videos that can generate an endless variety of playable action controllable worlds from syn thetic in images photographs and even sketches all right so here's a link to the actual Genie project as well as the um research paper and we'll go over um a bit

Of both of those so Google Deep Mind Genie Google deepmind introduced Genie an AI that can create playable video games from just a description or picture this AI called Genie learned from watching 200,000 hours of video game play online without Specific Instructions surprising ly even without uh direct training on game actions or descriptions Genie can figure out the main character in a game and let players control them in a these newly created game worlds this works through some advanced technology including

A lent action model and video tokenizer although it's only for research and not available to the public yet with some current limitations like generating games at a low speed one frame per second this development is seen as just the beginning soon people might be able to design their own unique Virtual Worlds additionally Genie was also tested with robotics videos to explore its use in creating models for robots the team behind Genie believes that this is an important step towards developing versatile

General purpose AI Models All right so we'll dive into more of everything we just mentioned in the second in detail here we have Tim open-endedness team lead at Google's Deep Mind basically posted saying I really excited to reveal what Google deep Minds open-endedness team has been up to we introduce Genie a foundation World model trained exclusively from internet videos that can generate an an endless variety of action controllable 2D worlds given image prompts so generate a playable World set in a

Futuristic uh City here's an example of that right here boom now Jim fan tweeted out saying this is my favorite paper in 2024 so far Sora is cool but Jeanie will actually be the backbone for embodied agents congrats Team all right and Tim continue tweeting saying rather than adding inductive biases we focus on scale we use a data set of 200,000 hours of video from 2D Platformers and trained an 11 billion world mod model in an unsupervised way Genie learns diverse

L actions that control characters in a consistent manner so he says our model can convert any image into a playable 2D World gen can bring to life human design Creations such as sketches for example beautiful artwork from senica and Caspian two of the youngest ever World creators Genie learned latent action space is not just diverse and consistent but also interpretable after a few turns humans generally figure out a mapping to semantically meaningful actions like going left right jumping Etc playing from

The same starting frame can generate diverse trajectories as you can see here mely opening eyes sore is really impressive and Visually stunning but as Yan says a world model needs actions gening is an action controllable World model but trained fully unup supervised from videos so how do we do this we use a temporarily aware video token gizer that compresses videos into discrete tokens a latent action model that encodes transitions between two frames as one of eight Lon actions and a mask

Git Dynamics model that predicts future frames all right you can see that diagram here and no surprise here data and compute we trained a classifier to filter for a highquality subset of our videos and conducted scaling experiments that show model performance improves steadily with increased parameter count and batch size our final model has 11 billion parameters you can see here from the paper figure 9 scaling results left training curves for different model sizes middle final training loss for each model size

Averaged over the last 300 updates and on the right final training loss for a 2.3 billion model with different batch sizes and we're going to kind of dive into the paper in just a second and kind of break it down in an easy to understand and digest way and also too this is really cool because Genie's model is general and not constrained to 2D we also trained a genie we also trained a genie on robotics data r21 without actions and demonstrate

That we can learn an action controllable simulator there too we think this is a promising step towards models for jgi consistent behaviors with the same action simulating deformable objects so this is leading into robotics as well and we're going to talk more about robotics in some future videos on this channel because we're seeing a lot of advancements in those fields right now all right so this is the genie project website I'll leave a link in the description below for this you

Can see different examples as well as read the actual paper here and we're going to dive into that in just a second you can see different sketches being animated as well as different drawings different pictures images Etc and as well as the robotics down here and also here is the genie research paper we're not going to read through each and every part of this paper but I will break it down into an easy digestible way basically breaking down what Genie is

How it works and what this could potentially mean for the future of AI AGI as well as every industry your personal life maybe if you run a business what that means for your business so you can actually start preparing for these advances so you're not just left in the dust and you're ready um when this stuff happens okay so what is Genie we already talked about this a lot but Genie stands for generative interactive environments it is a breakthrough AI Paradigm

Um that allows the generation of interactive playable environments from a single image prompt unlike previous models which required action labels for training such as open AI here when they are learning to play Minecraft with video right you can see this video right here they actually taught the AI to play Minecraft and I had this in my last video but um that actually required labels for the training Genie actually didn't require any right so how does this work one learning from videos

Without action labels Genie uses a large data set of 2D platformer game videos and Robotics videos from the internet learns fine grain controls and infers what parts of the video can be interacted with by observing the environment's Dynamics without needing specific action labels or annotated data two latent actions so instead of relying on labeled actions Genie use uses unsupervised learning to discover latent actions essentially uncovering the verbs that describe possible interactions in a video These latent actions EX example jump move

Left move right show consistency across different environments indicating that Genie has learned a set of fundamental interactions three creating playable Worlds by feeding Genie a single image it can generate a whole new interactive world this process is made possible through a learned latent action interface it can use various prompts including handrawn sketches text generated images or even real world photographs and then four contributing to AI agent training and they mention this in the paper Genie generated environments can be used to

Train AI agents offering a virtually endless variety of scenarios Laten actions learned from videos can be transferred to new unseen environments demonstrating potential for training more adaptable AI agents so as you can see a lot of different um companies now such as obviously open AI now Google uh and really just any AI company they're really pushing for that advancement of AI agents and we'll talk more about this in future videos open AI has basically said that they're working towards that who

Knows if maybe they have obviously they probably have a lot more advancements behind the scenes maybe they even have some agents in their Labs that we don't know about yet since they're rolling things out very slow right now but uh this Genie Tech and this new kind of way allows you to train these agents a lot more efficiently in different scenarios and it can be transferred over in new environments so that's really cool really powerful stuff technical aspects of Genie training

Details Genie comprises several components including a video tokenizer a Laten action model AKA lamb and a Dynamics model the use of various Transformer architectures optimized for both space and compute efficiency enables Genie to predict interactive video frames effect effectively to generality and scalability so Genie isn't limited to 2D platformer games its methodology is designed to be General enough to play other domains scalable to larger data sets Genie has the potential to leverage more internet videos to simulate even broader sets of

Real world and imagin environments and then three controllability and diversity so Genie showcases a high level of controllability where consistent behaviors manifest from the same latent actions across different starting images generated environments exhibit diversity and Rich Dynamics such as simulating basic physics or emulating Parallax in scene so that's one of the emerging properties it was actually able to learn and emulate Parallax and Parallax is that effect when you're basically scrolling let's say you're scrolling down a web page or you're scrolling

Down a a a scene or image and the front the back of the image say the the background is kind of staying the same while the front of the image let's say if there was a mountain here is actually scrolling up and I can show an example of that here potential impact and future directions so one empowering creativity so Genie opens up possibilities for individuals including children to create and interact with their own virtual worlds create video games the ability to

Generate game like environments without requiring professional game developing skills and this is kind of similar to what Jason hang said about who is the CEO of Nvidia about maybe us not really needing developers or coders anymore in the future basically where he said that back in the day that was one of the most important skills we wanted our children to learn whereas now they're building a future where we don't want our children to have to learn those skills because anyone can

Be a coder anyone can be a developer anyone can build a software anyone can build a game and this is definitely going to change the world it's going to change how businesses run right right now there's so many different SAS companies software companies and all these companies get a huge eida multiple uh say if they sell the company and they charge these huge profit margins and there's really not too too much besides like server costs and whatnot in terms of or

Your actual uh expenses are for providing the service providing the software and this is probably going to shake up the software well it's going to shake shake up the software industry a lot when you know companies are actually able to take either software inhouse replace different software because they can just spin it up with AI and um you know it's also going to change the gaming industry of course as well and pretty much every industry we know this but those are

Some of those direct ones that are going to get impacted from something like Genie Sora Etc uh second is Agent training by creative creating a variety of training environments on demand Genie could facilitate the development of more General adaptable AI agents capable of performing in in diverse situations and then three improvements and challenges so future work could focus on enhancing the model's resolution temporal coherence and interaction rates addressing hallucinated or unrealistic Futures in is another area of potential Improvement and also

Too I want you guys to think of the potential of this you could create a video game just from an image and we just saw Sora and you know I showed some of the new update videos here earlier in this video imagine taking some of those videos and using something like Genie or having open AI or sore or a different type of model that's basically the combined has the combined capabilities of both of them obviously this is the dumbest they'll ever

Be so by that time it's going to be even more advanced you could create lifelike images or videos from something like Sora and then use something like Genie or whatever capability it has to create a video game of it right so you could create video games of real life you could take a picture of yourself in your room and you could create a video game of your own world right so then it begs to question as we get more advanced as

These things start getting better and better and more realistic you know people are going to be using things like apple goggles or uh Apple Vision or whatever you know the the headset as well as like The Meta headset or different headsets that come out people are going to be ingrained and plugged into these Virtual Worlds these virtual realities and a lot of times when you see AI and uh all the AI that we're testing here we're testing it in these virtual

Environments AI is learning these virtual realities these World models learning physics such as Sora Genie learning physics it starts to beg the question okay well we're basically on track with simulation Theory and if you're not familiar with that we'll kind of go over some of this real quick with some philosophical theories so not going to give you a whole overview of um you know simulation Theory but basically it would beg the question well it would basically lead to the fact that

We're going to eventually be able to build simulations that are indistinguishable from the reality we're living in right and exploring the intersection of AI and reality so the advancements in AI Technologies like Google Gemini or sorry Google Genie and Sora which can create highly realistic simulations bring us to the brink of a new frontier where the lines between reality and simulation really start to blur and this races profound questions about the nature of reality and our place within it touching on

Some theories such as simulation Theory the goddess Spinosa and the concept of the universe revealing itself through Ai and these are just some things that have been on my mind I'm not saying that you have to believe them or that they're even 100% true these are theories right and these are things to get your mind going especially with all this Ai and the advancements it's kind of starting to bend reality at least for me it is right and I'm really into

You know researching the latest on AI obvious obviously and all this stuff so question is are we living in a simulation and indistinguishable simulations and this is something that Elon Musk has talked about many you know Sam wman has talked about and I've heard Sam wolman's opinion kind of like yeah we good chance we are but I mean I just what am I going to do about it right and that's some people's you know kind of thought about it um but

A lot of you know really rich billionaires Tech AI enthusiasts and all these people that a lot of them some people don't believe in it whatsoever um but some people are like yeah we don't really have much to disprove this Theory so and now that you're you know we see AI it's becoming more and more real so as AI generated simulations become increasingly realistic they could one day be indistinguishable from reality lending Credence to the simulation Theory which pause it that

Our reality might itself be a simulation created by a more advanced civilization or you know something else right and the philosophical implications so the possibility that we could create such simulations raises philosophical questions about the nature of existence and our understanding of Consciousness and reality technological singularity the advancements in AI could lead to a point as the technological singularity Ray kursell talks about the singularity a lot he's been talking about for a while definitely look into some of that where AI

Surpasses human intelligence and I've talked a lot about it on other videos and I'll continue to talk more about it in you know future videos but potentially allowing for the creation of even more complex simulations so really crazy stuff and the god of Spinosa if you haven't heard of this I'll give you a quick rundown on what it is and you know kind of how this could connect to Ai and what we're seeing here so few different points everything is God

God uh Spinoza's God is not a personal deity but the totality of the universe itself according to Spinosa everything that exists is the manifestation of God and nothing is outside of God Ai and Spinosa is God if we view AI as part of the universe's unfolding then the advancements in AI could be seen as part of the natural progression of the universe revealing itself akin to Espinosa's pantheistic view understanding the Divine as AI helps us to simulate and understand complex systems

It could be argued that we are getting closer to understanding the mind of God as sposa envisioned it where God is equivalent to the laws of nature right we're building this thing that's understanding language now it's understanding physics it's like it's learning the code of the universe right and that's kind of what the goddess Spinosa is where you know Albert Einstein believed in the goddess Spinosa as well as Elon Musk believes in the goddess Spinosa which he stated on Lex freedman's

Podcast I believe in Spinoza's God who reveals himself in the orderly Harmony of what exists not in a God who concerns himself with Fates and actions of Being Human Beings right and the the goddess Spinosa basically is the concept of the god revealing it it itself through the natur of the universe in physics right so let's say us creating AI is basically you know that nature of physics revealing revealing itself and um potentially if it's able to build simulations it's like

Basically building Another Universe like say if we're in a simulation right now it's like a simulation within a simulation and it's able to start building these simulations that are indistinguishable from reality it's kind of along the same lines right at least those are some different theories that have been kind of going on in my head so let me know what your thoughts are in the comments down below Universe revealing itself through ai ai as a mirror the idea that the universe

Is revealing itself through AI suggests that as we develop more advanced AI systems we are essentially creating mirrors that reflect the complexity of the universe itself insight into reality ai's abilities to simulate and predict complex phenomena could provide insights into the fundamental workings of reality potentially offering a glimpse into the underlying structure of the universe and philosophical and scientific exploration this concept aligns with both philosophical in inquiry as and uh scientific exploration as we seek to understand the nature of reality

And our place within it all right so that is pretty much it for this video going over goule Gemini Google uh Gemma as well as Google Genie and what that could mean for you know the shortterm as well as a longterm as well as what this experience actually is all right so let me know what your thoughts are in the comments Down Below guys I think this month of AI has been crazy it's only going to get crazier so make sure

You smash that subscribe button smash that like button let me know what your thoughts are in the comments Down Below guys also you uh have been noticing um this scrolling bar right here this is going over a few different free things I have for you guys um we have a free stride Community free AI stride Community with a bunch of free value in there so make sure to check that out it's stri community.com second is um our free Six-Day challenge that

We'll be hosting very very soon so it's accelerate your stride. comom it's a free AI challenge where we're giving a bunch of free resources bunch of free training this is stuff that people charge a lot of money for that we're doing for free so you can get your spot secure your spot in the future we may charge for it but um if you want to know how to leverage AI in your business as well as your personal life you know developing

Yourself Etc also too guys if you don't know I just published my first book it's a big one and it's jam-packed with value if you run a business if you need clients if you want to leverage AI if you can if you actually want to be the Chief Executive Officer of your business and not the chief everything officer of your business where you're doing everything and you have no time to live your life enjoy your time with your family do things

That you love then make sure to grab a copy of that book the digital version is down below uh clientacquisition code.com you can grab the digital copy of this book as well as um a bunch of free bonuses so go check out that web page you're going to love it I put my heart and soul into that as well as spent many many many hours writing it so go check it out we cover everything from Paid ads organic marketing content creation

Etc so prompting using prompts and chat GPT creating assets for your business how to set up your CRM all that good stuff speaking of CRM last thing is if you are familiar with go high level we have a custom built Master snapshot that we're giv away for free for people who sign up to go high level through our affiliate links so go check out G St level there's a bunch of free stuff in there it's 100% free just sign up for

Go high level which is the best CRM in the world you don't know about go high level just go to ghost.com there's a video explaining everything if you run a business you pretty much in my opinion you need to be using this so um that's my opinion go check it out you don't believe me watch the video do your research on go high level as well as um the the website explains everything about our snapshot that you get 100% for free

Which is people could charge thousands of dollars for these some people do but you would get it completely free and I hope you guys aren't too stunned or shocked from this video anyways guys that is it for this video let me know what your thoughts are in the comments down below smash that subscribe button so you can stay up to date with all the latest AI news and how to actually Implement AI into your business into your personal life so you

Can become a best version of yourself and I will see you in the next video keep hustling keep grinding and of course accelerate your stride guys take care