OpenAI's "World Simulator" SHOCKS The Entire Industry | Simulation Theory Proven?!
ฝัง
- เผยแพร่เมื่อ 20 ก.พ. 2024
- OpenAI's Sora is described as a "world simulator" by OpenAI. It can potentially simulate not only our reality but EVERY reality.
Use this limited-time deal to get an additional discount for all yearly plans with my code “MATTHEW” or link: hostinger.com/matthew
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Links:
Sora - openai.com/sora
Devs - www.imdb.com/title/tt8134186/ - วิทยาศาสตร์และเทคโนโลยี
![ใจเสาะบนเกาะนาคี - อ๊ะอาย สกุณาพร เสียงวิหค [Official MV] | ค่ายเพลงเป็นหนึ่ง](http://i.ytimg.com/vi/jbcmGQkfyOg/mqdefault.jpg)
![SORA AI’s Problems [And Solutions]](/img/n.gif)
Does this prove the simulation theory?
No. Simulation theory cannot be proven from inside the simulation. All we can achieve is credibly confirm the possibility.
No, it doesn't prove simulation theory. All it shows is that machine learning is capable of developing a highly accurate model of the world. One of the biggest mistakes you can make is forgetting that the model is not, and never can be, fully isomorphic with the system it is modeling
No? Not even close lol. Simulation theory is just like AGI. By the very definition of it you wont know when/if it happens and the constant videos about it are kinda just clickbaity. Its getting a bit old not gonna lie.
@@Mark-dc1suI’m never gonna say never.
No, it doesn't prove it, but it contributes onto building the theory, maybe we should wait 3 or 4 years more to prove it.
With all these SHOCKS, I’m starting to think AI safety needs an electrician. 😮
ROFL
yeah, his titles are a little "shocking" recently. im rather not watching them then getting disappointed. only trigger if there is a strong trigger.
I'm SHOCKED at how SHOCKING it is that this SHOCKS the industry!
Is this a meme or something? Wes Roth is doing the same. I don't get it.
@@MrTrialgodI follow Wes as well as Matthew and enjoy videos by both of them. It's just a meme on how everyone is using as many variations of the word SHOCK in their video titles these days. All good fun though
Shocking comment :O
@@MrTrialgodRussel Brand too 😂
Just a small correction: A 4090 can render a frame of a AAA game in less than 10ms. Sora won't even fit into memory and even if the 4090 would have enough memory for it, it would take seconds for a single frame.
Edit: What many don't realize about Sora and similar tech is that although it's extremely surprising to see it already, the computational complexity is huge. It would be impressive if it could render 1 frame per second. Then you would wait 24 minutes to generate a 1 minute long video assuming the typical 24fps of a video. In reality you would probably wait multiple hours unless you are Sam Altman with access to higher computational speeds.
Sure but we've gotten down to sub-second image generation with SD in about a year and half. We're seeing that if you tailor your LLMs to a specific task you can get them running very quickly on consumer-grade hardware (think something like moondream for img-text). There is no reason to think as far as I can tell that we're anywhere near a fundamental limit on speed. Imagine sora-mini models that just contain enough of the latent space to make a doom-like game. Or even a process similar to shader caching that generates enough of a game to start, and then just computes in the background. And that's not to mention alternative architectures (again, like we see with stable cascade v stable diffusion or mistral vs llama)
I don't think many folks don't recognize the truly insane amount of compute the OpenAI has at it's disposal, it's just that's not really stopping anyone from exploring ideas from the "eventually we'll get there" perspective. Because traditionally, we have, and surprisingly fast.
@@veqv Agreed, I wouldn't be surprised if we have a better Sora model in 10 years that can generate a 4k image in under a millisecond.
Was just pointing out that currently it's nowhere near that speed which was a claim in the video. Current SD gens with ms gen times are pretty bad.
Sora is still a surprise and the implications of an ability to create a consistent internal model of the world are far reaching. Imagine fine tuning the base model for a robot to teach it a consistent model of the world AND itself. It should be able to navigate through the world quite well. A model that understands physics and can accurately predict the next frames is also key to autonomous driving. The current architecture needs modifications for sure, e.g. I believe a recurrent component is necessary to really be on the same level as an animal. Some architectural combinations will lead to amazing results which Sora displayed quite well already.
I honestly can't follow his argument about this being any more efficient than Unreal Engine 5.
If there is an optimal explicit solution for a problem it's by definition superior to the implicit, bruteforced solution.
The Kolmogorov Complexity of UE5 is probably less than a millionth of that of Sora, and it can probably still be drastically reduced.
At the limit, deep-models like Sora will always depend on explicit models or will need to become explicit themselves.
Also in terms of search-problems ("innovation"), only explicitized models can reach all solutions within the search space, as even the slightest bias eventually leads to a recursion.
And may I remind you, that many people in this world still believe the true binary-size of the universe to be equivalent to that of one elegant explicit world-formula.
Chips are being redesigned as we speak, made specialized for AI
@@the_curious1 AI is accelerating at a double logarithmic rate. I give it 4 years.
Okay, plot twist. Sam Altman has reached an agreement with the machine world, and they have agreed to wake us all up. So now, he is slowly deploying the timeline that makes us aware of the matrix. And the moment we all fully grasp and live with the technology available, we get woken up.
I hope they let some of us stay in simulated worlds, but upgrade the UI so we have more control over it. Like, "Stop. Back me up a minute; let me try that again." Or, "That Putin NPC? I want him to have stage three metastatised bone cancer twenty-five years ago. Back me up to that and...oh, wait. Hiccups. Give him incurable, persistent hiccups. Okay, back me up to *that* and restart."
@stevejordan7275 that sure would be cool. Unfortunately I find the timing of these lifelife simulations, climate change, trumps antichristlike attributes, and ufo disclosure troubling.
Nope. They’re just starting to put us to sleep. I imagine we’ll all want to wake up soon.
I've noticed some popular streamers saying they can't see any positive outcomes from SORA, only negative (scams, manipulation, job loss, etc.). I don't think they realize this is more than a text-to-video model. It is a world model that has learned a lot about the physical world. There are a LOT of problems that can be solved using an accurate world model like this.
This is the first thing I tried to do. I also wrote in 2004 that this is how you could wake up a conscious AI. Wish I could find all the notes I wrote on AI for 20 years.
I love how AI Explained sets the standard for high quality content and simple descriptive titles and is twice as popular as any of these high effort, entertaining and informative channels which feel it necessary to SHOCK THE ENTIRE INDUSTRY!!!!!!
I was subscribed to AI Explained but with custom notifications and subscribed to Matt's channel with "all videos" selected. This video and reading your comment was the last straw convincing me to swap the subscription settings for these two channels. Looking at the videos that AI Explained has produced recently, it's clear his goal is to put out quality content. It's also becoming increasingly clear that Matt's goal is to become the Mr Beast of AI videos. The endless barrage of clickbait titles, low-effort, bombastic content, and his obsession with simulation theory are almost enough to counteract the useful tutorials he occasionally puts out. 🙄
Great comment
Matthew is only rolling with a meme, trying to appease people who cling to it because he's done it a little bit in the past, and Sora really is shocking.
I count 6 people in this very comments section threatening to unsub due to it.
Not a good time to roll with a meme not gonna lie..@@mattmaas5790
I actually looked into this a few days ago because a guy from Meta commented on this. The guy said that the way OpenAI is going about this is wrong and that its impossible to account for things like randomness, chaos, physics, etc when generating worlds using pixel by pixel methods because what it does is predict what the next pixel should be but it has no understanding of how the things in the image, like physics, work. So you end up with someone taking a bite of a cookie and the cookie remaining complete. So apparently meta has this different model called JEPA (there are two variants) that uses computer vision to analyze the image, text to understand everything in it, and then turns it into video
From perplexity,
"JEPA models, like I-JEPA, understand images and make predictions based on high-level semantic information learned during training. They focus on predicting representations of image parts rather than detailed pixel patterns. Sora, on the other hand, generates pixel-level content based on patterns it has learned, simulating actions and interactions within videos.
In essence, JEPA models are about understanding and predicting, while Sora is about generating and simulating. Each has its unique applications and strengths within the field of AI and machine learning."
The universe is quantum. It has both spatial and temporal resolution depicted by the Plank constants (length, time, mass, temperature). Those are the quantums in our world. The question is, can a universe create a sub universe at the same resolution as the original? If yes, then we are most likely in a simulation.
What you are seeing with Sora is an image generator creating images in succession with enough temporal relation too fool the user that it is a real 3d world model. However, the actual world does not exist. If the model output pixels "looks" forward at some distant mountain, then pans down to look at the ground so the mountain is out of view, it no longer exists in the model. When the camera pans up again, the mountain could have disappeared or moved. The whole thing is basically a fever dream that is just good at connecting pixels between pictures. People are too hung up on what images it can generate without thinking about the consistency. It also needs to actually create the world it simulates in some other form for it to actually be running a simulation. This is just pixels on steroids.
You can clearly see this btw in the Sora demos where part of the image pans out and into view again. The AI has dreamed up something different in those areas. Same with occlusion, the moment those objects are occulded they do not exist. The model fools the user to think that something was behind it by generating something that looks right but isn't really right if it was a real 3d model. Hence people will disappear when they move behind total occlusion and we expect them to exit again at the other end.
For these AI models to actually have a world simulation they would have to first create a a real 3d model that represents entitites and move those around before using that as input to an image generator, but even then it will struggle with consistency as it has no actual storage of texture and shapes as it has to dream those up on the fly for every time they should appear in the image.
Also, that land rover is simply an example of taking a source video and changing elements via a prompt. They used video game videos as the input and changed the cars and setting etc.
Are you saying you are not SHOCKED ??
I think Sora used Biomes. Remember back in February 2023 when Global Illuminations was bought by OpenAI?
Their WebAssembly game framework would have been much faster than Unreal Engine.
SORA currently takes a long time (at least a few minutes) to generate video based on text input. So may be a while until it can be used for video games because it would have to render output based on user interaction in “real time” with only a few millisecond delay. Even ChatGPT can take several seconds just to generate a text-to-text response. Faster model-interface architecture combined with more powerful local device AI chips could get us low-enough latency.
This. I think Matthew forgot about the output delay when interacting with the simulated world.
Nope, I think you guys are not realizing that they are speculating on a refined video game level product. There is no video game that looks as good as these simulations. If they reduced the complexity of the world simulation to be just beyond where video game graphics are right now, it could be real time. Someone did real time playable GTA simulation already, it's on youtube. Basically this takes a few minutes, but its simulating realistic reality. NO video game looks as good as that chinese dragon new year festival. If they reduced the model, it would work, other reseraches have gotten GTA V to work this way already.@@zodiacblue9312
I appreciate your capsulation of content. It causes me to predict various other potentials. It's like your forecast is one of many possibilities, and your sharing of it stimulates the mind to read between the lines, and/or envision yet a different future.
keep in mind the sora Minecraft is only a video that looks like Minecraft not a game. You could however use a depth map etc game output as input for sora. Sora has internal consistency and approximated physics but it's always randomly generated imagery without the control of a game engine
Don't know if it's possible, but could be one hell of a way to sandbox AGI / ASI (in simulation)
Do we know how long does it take Sora to render those 1 minute videos? I'm not an expert on the subject but considering it takes almost a minute for Dall-e to create a single frame, rendering these videos must take so much time. So the question is, when are we going to reach that level of tech which would enable us to create these videos real-time.
OSS models are much faster then that with LCM and TensorRT.
Someone on reddit achieves 150fps gens.
I don't have internal details on Sora but my guess is generation time will decrease rapidly
Not "time", but calculation power. And they certainly use the equivalent of like 100 GPUs at once, your it's not usable by your common man.
A guide on integrating the new Zed text editor with Ollama would be great. It may be as simple as redirecting traffic headed for OpenAi’s GPT-4 endpoint to a local port there Ollama is being served, since Ollama supports the OpenAI API spec now. Should be a quick guide to make and very practical.
"The creators binary" - nice one
I'm not sure if anyone else has already mentioned it, but this reminds me a lot of the movie The Thirteenth Floor (1999). If you've never seen it before and you love Sci-Fi movies, definitely give it a watch.
Welt am Draht (World on a Wire) is a great German Miniseries from the 70s that explores simulation theory.. highly recommend
Yes, last time I was watching it, a copy could be found on TH-cam. But even if not, purchasing and watching this masterpiece is a must. Not only the story is great but the acting and camera are fantastic. For me it is R. W. Fassbinder's best piece.
Plot twist: our personal views are already simulated by Sora v92
Imagine if in 2040 it’s widespread societal collapse and then all of a sudden you wake up realising your whole life was a prank. Your little bro put you in a sim while you’re sleeping.
If this happens to me. God why
I can't wait for these dynamically gemerated video games to be paired with a direct neural interface. And what if the outputs were directed to more than just pixels, but also to manipulation of 'hardware'? Maybe on a microscopic level. That might be interesting...
and SHOCKING !!
Ok, I am starting to notice a trend. You and Wes Roth are using "SHOCK" more than ever. Are you both using the same AI to write click-bait titles?
I get what you're saying, but what kind of headlines are you supposed to write for videos covering brand new discoveries that shake the foundations of human understanding?
all in the memery
I've noticed clickbatey titles becoming more common again across TH-cam, wonder if it's an algorithm thing
Pretty sure people just have him a bunch of shit and so he changed the titles but they underperformed on view metrics relative to clickbait
You won’t BELIEVE what Sam Altman said next! 5 things that you probably missed during his speech that SHOCKED the industry !!
Considering that Sora is not just about simple video generation and creation, but rather simulation on top of physics simulation it is possibly possible to create a Matrix. Sora, ChatGPT and a VR headset.
Sora learned to intuitively understand physics the same way we do. We don't need a scientific understanding of leverage to understand that. Putting your back against something and pushing with your legs is more effective.
I'm SHOCKED, you're SHOCKED... EVERYONE IS SHOCKED!!! And the fact that everyone is SHOCKED is MIND BLOWING!
doctors and teachers don't want anyone to know
TH-camrs hate that you know this secret
Your comment STUNNED me with this AMAZING insight, you captured all our feelings with JUST ONE SIMPLE POST!
There is an original Star Trek episode where two worlds _Simulate_ an ongoing, 500-year old war between them. They then - individually - impose calculated casualties on their own citizenry. Assuming a human derived GIGO/SISO based AI learning structure, this will be our future.
"A Taste of Armageddon."
Now imagine each simulation being a simulation of another simulation, a chain of worlds. And the beginning and end of the chain being connected, so that none of the worlds is reality.
this video needs to be seen by a few billion more NPC's such as myself. You are literally one out of like 4 people talking about this. No one can call us simulation theory folks nuts anymore. I'm scared. I was worried before Sora, now I'm Neo waking up in the pod scared. Keep up the thought provoking stuff brother.
First I'd like to say this is exactly the route that I was hoping AI would develop in. Fantasy simulation games are my absolute favorite version of video games and I'm very excited to see how the future will go.
Second simulation theory also gives us the opportunity to experience the universe at a significantly lower cost than how the real world actually functions. Instead of using the complete amount of energy necessary to convert one chemical to another or to generate one specific type of resource, such as the amount of energy necessary to compress hydrogen into iron, a simulated universe would be able to skip the necessary energy by just predicting what the end result would be and plopping that into existence making for a significantly higher efficiency for everything.
Perhaps it can help us build a working warp drive. Having said that, I too can’t wait for these new systems to be used to create alternate realities for me (us) to explore and experience.
The fundamental difference between reality and simulations/stories is not sensorial, but structural in that the simulations/stories are made for people, designed, the real world is not and that in a way is the incentive to pursue the simulation. To see the difference you don't necessarily need to peer into the designer, but inspect the complexity and behavior of the phenomena. Games have quite clear rules and balanced challenges, the world does not.
Thank you Matthew for making a video on this topic. I have been in a research / content hole regarding simulation theory since the release of Sora, and it crossed your mind as well. I am pretty convinced, and regardless very excited about what Sora will bring. PS dont let the shocking meme bother you, people will meme about anything and we both know this stuff if shocking
SHOCKING advice
I share your sentiment and excitement. So, sure, we no longer need old and forgotten IP to resurrect an old game. Neat, but more importantly, we can synthesize historic places and events. Imagine standing in the crowd at the Gettysburg Address? or taking your old family photos and grandfather's stories and create that world again. The idea of (movie) Strange Days' SQUID memory and experience recorder is no longer needed to share experiences.
BTW, I think your titles completely reflect your excitement and personality. Complainers don't HAVE to watch your every video. Thank you for your content. I'm always excited to see a new video that sums up the latest ai leaps.
It's impressively good, but obviously still needs further improvements.
Such a world model will likely be essential for an ASI to evaluate decisions. This is just another step towards that and along the way, we will also get awesome products for users.
Also for those that watched the tv-show Supergirl: There is a (so far) fictional technology presented there, where users can get any fully immersive VR-experience they want on demand. That is becoming possible quickly. I'd estimate that we have something very similar within a few years where the visual quality is actually great.
As someone who's pursuing a dream of becoming a 3D Artist & a Game Developer I definitely don't see my dream getting crushed
15:19 - OMG, I now realize that since it's able to simulate all realities in a way then maybe one day the AI would be able to even show our lives as if its been recorded? (so a person is going to show the model how he/she looks and then the AI model would be able to tell in which realities this person exists and maybe to show him/her all of his life from birth to death?) that's way too wild and crazy if it would turn out to be true, also it's scary to think about the realization of it, it's getting into science fiction realm in a way that become real with Sora abilities, I still cannot believe we have such crazy advanced tech that is like a time machine already, it's not like we're in 2050 - 2070 or so, it feels like a super advanced tech from the far future.
If the ai industry keeps progressing at this pace, we will have to rethink everything because these models will output so many discoveries in all the industries that we won't be able to keep up.
I would say it proves it is possible to simulate a world or maybe even a universe. However proving simulating a world is possible doesn’t mean that we are for sure living in a simulation.
A bit hard to believe that is lighter than UE5, I want to see it running locally.
Not just video games. Once you put this in VR headsets you have a portable holodeck.
Its not the pixels that get simulated, the pixels are a connection that shows a representation of the calculations.
I am paralized from the shocking videos i watched this week!
If this happened in 1977,I would have become the ARCHITECT in 2066💪
So, Sora could generate game worlds on the fly while we control the camera or the player with an controller? ... ❤ to all hard working game developers out there.
I am confused when you say the video generator is faster. Sora probably takes massive processing power compared to a typical GPU. I think GPT-4 for example uses an entire cluster of very beefy GPUs. A 4090 could probably run games that look as good as the Sora videos but obviously designing all the interactions and models would take a long time, but it would also be more realistic.
It takes an incredible amount of GPU compute to train a model, but inferencing (generating an output) once the model is already trained can be done on hardware several orders of magnitude less powerful, like a laptop GPU. There are several free and open source methods for inferencing GPT completely locally on normal consumer GPUs
I agree. Like how long does it take to create a 1 minute long video. It's not creating it in real time which would be needed for a game.
Edit Games need to update every frame, this tech figures out a whole minute and puts it all together to create a video. It's completely different.
Edit 2 but I could see something like this being used to create worlds for unreal engine.
@@SpirusFilms You still need beefy GPUs with lots of VRAM for inference though, I think it was mentionned in a twitter thread that 1m takes 1h on consumer (4090) hardware with Sora
@@SpirusFilms I am saying GPT-4 inference takes an entire cluster. Training takes a super computer.
Imagine games create by "live videos" with endless expanding content. That would be then more or less the matrix
Look, if we were in a simulation there would have to be limits imposed on the world that we would notice. Like a maximum speed or a minimum particle size. And to avoid detection those particles would have mandatory vagueness. Meaning, I don't know, let's say you could know where a particle is but not how fast it's moving. Or vice-versa, I'm just spitballing here.
And even if it could work for earth, it would have huge processing problems anyplace too much mass gathered. Calculations would slow the CPU to a crawl and make everything seem slow. Pretty obvious.
What Matt said about things not being rendered until we look at them? If that was true things would be blurrier at the edges of our vision. So with just a little logic I think we can totally debunk this whole "Simulation Theory" thing.
Diffusion literally calculates ecery oixel 😅. It uses pixel noise to begin diffusing and in doing it is determining the colour of each pixel, refining it each time until the model agrees all pixels are the best match for the input.
I am unconvinced that a modern video game is more expensive compute than an LLM at even as low as 30fps.
It depends what level of complexity you're after. Things like ray tracing, fluid dynamics etc are extremely expensive to compute traditionally. A stable-diffusion style of weather-forecasting recently was shown to outperform the manually-written code.
Any idea when this level of sora will be avalible for us? I have dozens of ideas to worlds, games, and movies i want to create. Cant wait til this becomes available!
if i get a chat gpt plus subscription , what is the context window max, is it 32 k ? or more ?
Yeah seriously, the industry must shocked out with every little announcement causing such shock..
I wanted a console that could create any game or dlcs and this is just the thing. If we create a game/simulation realm and each different realm can be traveled between then that would be the multiverse and ours could be left behind or kept decent and we could go through the differences and then come out understanding and sharing with the "real" world or maybe heaven. Is that what this is? Maybe thats what this is, a growth of a difference that interconnects and reveals the hidden realities.
Wow just over 20 years from Nick Bostrom's " Simulation Hypothesis" and we're almost there!
I'm not scared, I'm excited. This actually makes me feel optimistic about the future.
It's as if all those wonderful promises are about to come true.
I will definetly remember this month...
Crazy time. To witness this...
It looks REALLY good but not perfect. I saw a video from Sora were a guy vanished into nothing while walking down the stairs in a broad view shot. He almost melted or molded into the environment. It was only noticeable because I was concentrating on the one person in the shot. If I was focused on the center point of the frame, I would've missed it.
I am PUZZLED about this reupload.
Pretty cool stuff!! I'm a big fan of old, old TV comedy show " The Dick Van Dyke Show"... I can see this tech being able to train on every episode in the historic library, and then render endless new episodes from scripted prompts with new plot lines. I mean endless!! You could do this with any show. You could probably even mesh old shows together. Simple Simon stuff... Months away from this I'll bet.
So the road is generated in real-time as you drive, which sounds kind of like the observer-effect in quantum physics.
One obvious problem with Sora is that it is largely a data driven physics engine, a data driven life engine is also needed to make a simulation that makes sense. Perhaps they are working on that too. Sora is still finite, and is only simulating as much as it knows, and only as well as it knows, and beyond that it can be creative but only speculative.
Can't wait for Mario Forever!
Thank you for your amazing content. We're literally going to be training EdenAgi using your entire video library!!
Sora might use an AI agent for managing layers to handle consistency with "remembering" objects behind other objects? 🤔 Thanks for finding all these nuggets among posted responses.
With realistic simulation, and realistic AI characters in that simulation - those characters may also ask if they are in simulation... Are they real?... Especially if their "conclusions, results and output" are used in our world, and influencing it. This is so interesting! :)
...and SHOCKING !!
I think that this technology will not be used to generate and play video games in the way you described. Not with consumer hardware and not real-time(even with huge amounts of computing power). Yes, it simulated worlds, but it does it on a big clusters of GPU's and probably generating a frame (one camera shot) every 1 hour or something. Also, at this point we have 1 minute video length limit and that is not coming without a reason.
Sora is essentially doing exactly what your brain does when you dream. This will change the world beyond what most understand. Sorry to the people that’ll have to get different jobs but it will be for the better good of everyone. It’ll be a shock like going from riding horses to two years later taking a rocket to space. I don’t think people comprehend this.
Just as you apparently don't need conscious beings to write code, you don't need them to "simulate" worlds. Some basic capacities like Turing completeness might be enough. Conway's Game of Life could (eventually) come up with Sora and it's "simulations", so can ordinary physics.
My quotation marks express my doubts about the notion of "similarity". Creations don't need to look like their creator, for some version of the Standard model's game of life may have "created" us without pre-conceiving us in some meta prefrontal cortex, like the one we use when we think we "create". No white beard, nor teenagers in the basement required at all!
The pace of change of what OpenAI is revealing is impressive and a lil scary ! 😅
What happens when you embody Sora into a robot with full sensors so that, instead of just understanding through videos and Unreal engine, it gains it's understanding directly through the real world, experiencing it in real time?
I‘m not getting the „generating in real time“. Open AI hinted that one sora video takes up to an hour to generate.
Wow incredible. What's next from Open Ai, Using it's horsepower to create the schematics for a holodeck. Enjoying your show's. Mr Berman
If we can live in the simulation, there is no need for us to leave the planet.
Sweet baby Jesus, between Matt & Wes shocks every 5 hours. My soul can't get a break!!
Could this be the reason we are seeing companies doing mass layoffs/firing in the video game sector?
I am seeing where game level design could be served here.
yeah, I can see model placement. I think a great question would be can you convert it back to files an unreal editor can open..
14:42 this has been proven already with the light particle/wave. The experiments showed that light is "conscious," and it changes from a wave or particle depending on if it's being actively observed. This is real.
Basically atoms collapse when we aren't observing them, so life is like a video game engine only rendering the parts that it needs to when it needs to. Mind blowing stuff
Is there enough memory in the world to really scale SORA to create full length films and video games with it?
Please dont become another AI youtuber who is SHOCKED by everything, I actually like your content but if you start doing this regularly I'll stop watching.
Embrace the meme
Brother at some point you need to humble yourself and just accept it lol.
The number of ( very fairly) critical comments I see you completely toss aside is insane.
It also seems to be increasing a noticeable amount..No sense throwing away your channel for a meme.
I am clearly not the only annoyed one thinking of unsubbing just in this very comments section. That should worry a creator, not cause you to double down.
@@matthew_berman
@@HistoryIsAbsurdworst part is, it's not even a damn meme. It's clearly intended to be ironic, but if you force irony you're not doing it ironically, you're just being stupid AF
This has changed The Entire Industry forever!
I aim to use Sora for a VR-enabled Star Trek series, allowing me to navigate among the actors.
I called out realtime ai generation games coming soon around 5 years ago. Fun to see my prediction playing out in real time ;D
Its not though... No one knows how this works and they have no proof with their claims.
@@helix8847 not yet, however it’s pretty clear to me that this is the direction were headed, and fast
Dude like my mind is completely blown like is this real life I’m experiencing right now or what!!!!!!!!
What do strict physicalists and materialists think about the implications?
Matt did you see Google's Open source model Gemma ~ It's Crazyyy
yep!
3:43
The complexity of a model is directly proportional to the complexity of the data.
This means complexity is computation dependent.
It’s doesn’t exist independently.
Therefore, Complexity is computation limitation.
Let’s assume that you have compute power to map this universe.
You started your journey from this planet and proceeded outward.
After an arbitrary period you finish and came at the point from where you started.
But now you don’t know that the point of your origin equal to yours current point.
Everything has changed there.
You discovered a earth like planet.
They will denote you came from previous big bang kinda event. From a previous universe.
But for you, you will perceive this as a parallel universe.
Computation limitations in Mapping of a sphere creates artefacts such as linear time and parallel space.
For a infinite precision turing machine there is no time.
For like a neural network or human brain low precision computation creates time and parallel reality in an inversely proportional manner.
111GB - 13:51 - Ain't that number symbolic due to it having 3 ones? coincidence or not?
Just about that Minecraft thing... Why is it, like loads of Sora's generations, in slow motion?
These 1min videos take a datacenter of GPU's to generate. Highly unlikely your desktop GPU will ever generate 60+fps using SORA. Probably closer to 0.1 frames per minute at best.
Is this a re upload . You already made a vid abiut sora
Unreal engine open sources its code so creators can edit the system to help themselves. I'm still new to game dev, but I can see OpenAi scraping the code to emulate the system.
Thats not the way it works... lol
Didn't you already made a video about SORA?
He did.... lol
@@MustardGamings This is a focus on simulation theory, with speculation from AI scientists and Matthew as well. I'm glad he made it and brought those simulation theory tweets to my attention. Devs is also on my mind.
that's basically what architects were able to do in inception
The compounding effect of AI learning is in full motion...
I think living in a Matrix would be amazing….. if it wasn’t censored. But OpenAI and all the other big companies seem to love to censor everything. Like imagine living in some pure Matrix reality but large parts of that reality was censored? That would be pure hell in my opinion.
If this would be proof of simulation theory, then we would see glitches everywhere, which we don't see. On the other hand, we do simulate our reality in our brains all the time.
Simulation theory is flawed beyond belief.
I don't think Sora is available to the public yet 🥺
This is so absolutely cool. Strap in folks, its going to be a wild ride.
How long until Sword Art Online is released? For me this is one step further to a Game with a immersive gigantic game-world with almost alive NPCs
You can play any game you want, just give them the rules you want to play by. As long as it's not a spicy game of course.