OpenAI Sora: Beauty And Horror!
ฝัง
- เผยแพร่เมื่อ 30 มี.ค. 2024
- ❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com/papers
📝 My Master thesis on fluids, with source code: users.cg.tuwien.ac.at/zsolnai...
📝 Paper/poster on fluid control, with source code: users.cg.tuwien.ac.at/zsolnai...
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Bret Brizzee, Gaston Ingaramo, Gordon Child, Jace O'Brien, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Putra Iskandar, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers
#openai #sora - วิทยาศาสตร์และเทคโนโลยี
4:06, the specular may be a bit high, but the light scattering through the leaves - still impressive.
Es gibt immer Luft nach oben.
there is no point in judging it as a CGI rendering and comparison, there is no real physics there, just replication in a "brutal force" way
Uhh, I think the AI generated a non-Euclidean house💀
Lovecraftian :)
try eliptic curve fourrier transformations of matrices in elementary systems of equations with variables of functions to the topology of the space
Welcome...to ai.
I want to see a video of an FPS game in hyperbolic space
Lol yeah
Around 4:03 the front legs of the elephant mess up and essentially switch locations in a mind-bending-physics-breaking-biology-wrecking flinch
Wow, nicely spotted!
Interesting; this also happened to the lady in the red dress in the video the OpenAI team put out demonstrating Sora's release.
Wow that's actually a neat transition
This happens in many of the Sora videos, like the woman walking with sunglasses in the street and the puppies playing in snow. The video of the African family does weird things with their hands, so not perfect but very close. Compared to about a year ago with the will Smith spaghetti video it seems decades ahead.
Sora may be the greatest lucid dreamer to ever exist at this point in time.
Sounds like an insult to lucid dreaming...
@petrac.1739 You mad over here for no reason😂
@@petrac.1739lucid dreaming is bad for everybody’s health
can you elaborate ?@@edupe6185
@@edupe6185It's not
With the house tour example it does actually seem like it _is_ making things up as it goes. It stays mostly consistent with whatever is in-frame at any given time, but it makes nearly a full right-hand turn and ends up somewhere that doesn't look like the same place as the start of the clip. Even from room to room, there are some inconsistencies in perspective that leads to e.g. the floor not appearing planar.
Also, there's a bed in the kitchen
Yeah, I have to take issue with his insistence that the AI "really has a model of the house itself". Sora may have much less consistency issues than previous AI video models, but they still clearly exist (especially in those "blooper" Sora clips, such as the weird chair at the archeological dig). And so many of those fly-through clips have a crazy amount of scale and perspective shifting that makes no sense in a true 3D context. On a fundamental level, it's still doing the same 2D "autocomplete" as prior video models - there's still no true spatial understanding.
placing a couch like that, right in front of the tv is VILE
I don't really like this youtube for this sort of reason. He rarely seems to approach things objectively, applying criticism even when it is obviously appropriate. It's more a channel where he just confirms his biases, and generates hype. Pretty misleading given the title of the channel.
Yeah, and those are small issues. If this was a quick scene in a movie 90% of people would miss it completely or just... you know, not care. These will be easily fixed.
What i'm dying to see that nobody seems to have tried yet, is a prompt telling the "camera" or "viewer" to pass by a mirror.
Without specifying "human" or "man/girl" or anything, it'd be interesting to see if Sora understands that the viewer itself ougth to be reflected in a mirror... And what would "it" look like when seeing itself in the mirror? Would it randomly assume it should be a person holding a camera, or something entirely weird?
I thought about this because in these videos with a lot of reflective surfaces, the source "behind the camera" is never reflected.
Really interesting idea. It's kind of like testing animals if they recognize themselves in the mirror but with a twist
I`m curious about that too! There is only 1 AI video about a train that eventually moves trough a tunnel. You then see the reflection of a woman holding a phone, filming the scene.
My assumption is that it would just choose randomly each time, unless you specify things.
It would probably just show a random person, like the woman filming in the train video
The result would be highly contextual based on the specific wording of the prompt in relation to the training data.
Damn, My grandmother at age of 83 is still alive and she was living in a old mud house at her teenage with no technological accomplishments, and AI is achieving these accomplishments now, God knows What AI will evolve When I become old.
fr. the fact that it’s growth will be exponential is kinda insane to think about.
This dude’s dead consciousness reanimated via AI in a live audio video format will probably be telling us about a new incredible AI technology that can kill the user, but, hold on to your papers fellow scholars, because it can do this without the user even knowing! What a time to be alive!!!
It will make sure you won't get old. In good and bad ways.
My grandmother passed away last year. She was able to witness horse-drawn carriages and GPT3 during her lifetime.
@@carlosrivadulla8903 wild.
ah yes, 5 right turns means it really understands the space it's showing/s
Come on, that's not really an accurate solution for Navier-Stokes. You can ask a decently good artist to draw you a whirlpool too. The result will most likely look convincing, but that does not mean he understands a squat about fluid dynamics (or light transport for that matter) - he's simply taught with patterns to a certain degree. And just like he can recognize the patterns, he can also CREATE them. And since your brain is ALSO taught on similar patterns - you see it as a "valid looking non-laminar flow of liquid". A neat trick for movies, but definitely not something you want to use in CAD/CAE...
So it doesn't understand the physics right? It just determines how to color each pixel based on all the videos it has watched?
Also human design has a purpose but it’s like it simulates and mix what it has seen but not understand why which can be problematic. I think a lot of people going to be disappointed when we try more concrete things.
Don't think it's that simple, just like LLMs are not "just predicting the next words".
LLMs need to learn the concept behind the words before being able to predict the next word.
In the same manner, SORA needs to learn the concept of what is an elephant before being able to create one made out of leaves.
Same goes for physics, and everything else, really. Hence why OpenAI is calling it a "World Model" and not just a "Video Generator".
I don't think this is just propaganda from their end, the successor to this thing will probably be able to generate full virtual 3D worlds on the run if it was instructed to do so.
@@luisfonseca9045 it doesn’t need to learn the concepts to do it. Better to learn how these works to understand the limits. Even if we put hundreds of millions in cost to train them it still can’t calculate numbers better than a calculator. It struggle to break down words to letters because it’s trained on words. Sora forgets objects when they are hidden and changes them. Sora does wrong numbers on legs on ants. It’s a lot of challenges and there are not answers on everything. But it’s not all or nothing, it’s great tools but we can’t predict the future.
shush, it's ethically wrong to imply that humans might be better than anything, especially AI
@@luisfonseca9045 Do you have an example of an LLM that can “learn the concepts behind the words”? All I’ve seen are LLMs that are better at attention (relative importance) and the _relationships_ between words, and are trained on larger datasets, but none that actually know anything. Is there an LLM out there that isn’t just a really good text generator?
"What a time to be alive!" -- now it is only true for those in power and control of these things. For the rest... not so good of a time.
Computation will hit a road block, even Sam Altman said they are expecting Nuclear Fusion (which is always only 30 years away) to help them across that road block.
Just in the news this week, Microsoft is dumping $115B on Stargate for OpenAI and Mistral, and Amazon is putting $150B into their own AI data centers for Anthropic. The compute side of the issue will probably to be resolved by the early 2030s, if the timelines hold up, although I'm sure this will only be the beginning of AI related megaprojects.
In terms of power generation, nuclear fusion isn't really necessary, except to bring down the price of electricity production, and I don't think it's certain that it will ever be a viable energy source, though the possible benefits are obviously high enough to justify the research.
In the meantime, Microsoft and OpenAI are already lobbying for fission plants in the shorter term. The timeline for their construction will be a lot faster than from any government initiative, since the funds will be readily available and not held hostage by "democracy". They also have the skilled labor and engineers on hand to design these plants, and the profit incentive for completion will be very high.
@@GeoMeridium Capital can only take you as far at the implementation you are utilizing, but all the innovation has come from years old transformer papers google published. I don't think this will be solved with the current implementation. Only time will tell what comes of this. Thank you on the update tho. :)
Yeah, but what's going to be accomplished before that roadblock? If the roadblock comes in 2035, even 2030, the computational advances will still be big enough to give us some crazy stuff
@@gunzor8717 you can polish a turd all you want, at the end of the day, it's still a turd. :D
@@juanmacias5922 Google played a big role in making transformers but to credit them with all the progress is kind of asinine. Tons of new things have been discovered since then.
"When everyone's super, no-one will be!" ~Syndrome, The Incredibles
The next step will be of a Clau3 Opus level AI looking over those images frame by frame, critiquing the errors, and feeding that back into the model as synthetic new data. This problem will be solved effectively over night.
Ok that house video is definitely non-euclidean. The kitchen goes into the foyer where you start out. Funny how I see that just as you say it doesn't make things up. Still very impressive regardless!
"What a [terrifying] time to be alive!"
If I've learned anything from this channel its to look for and appreciate the reflections and refractions in A.I. generated work.
2:24 I have never seen such a remarkable blending of VW, Mitsubishi and Subaru!
The real horor is the fact that everybody will become a director.
Like everything good that this AI brings, the bad will always come, there will be people who will use it to create garbage content, fraud, span and a lot of other things, personally I think that we are not ready to face the amount of malicious content that is coming.
Every time something good happens, there's someone there to poopoo it
Crypto was a good idea until it wasn't. The infinite greed and research for immediate quick profit ruins everything.
i dont like paywals but here there should be one so that exact thing doesnt happen, or atleast not as often
It's obvious for me that you're also aware of the dangers, clickbaits and such. But i doubt we'll be aware all the time. For example, the other day i watched a video about a french guy who filmed Pablo Escobar. He told how he managed to do that, what happened that day. In the end, after reading the comments below, i realized he was totally dubbed by IA into english! I got stunned!
I know about IA, i know very well how to detect most of it, i know what this is capable of. But even with this ''skill'', i wasn't able to realize it, because i wasn't unaware of this as possiblity on a random video on internet. That's crazy! I also follow a page of old photos on Facebook. Great photographs, historic ones and such. But now, sometimes i can remind myself and ask to myself if what i'm seeing is a real image or one made by IA. From now, we will not be able to trust anything, not even a video, a very consistent photo and such. This is crazy and a lot of adjectives. lol
Yeah I mean, it's not like we already have garbage content, fraud, spam, and a lot of other things.
sooo ... with the enough time, the ai will be able to create an entire videogame only with the words instructions ... and with even more time, it will be capable of create an entire simulation. What a time that we are living right now
Yeah the job market is going to fall apart and the difference between the super rich and everyone else will become larger. The ones that are replaced last are the heads of companies.
@@spumbibjorn This is the only sane, non-hysterical comment that I've seen on this video
@@benjaminsandeen9241 Hopefully it all turns out alright, but whatever happens, things will get shaky for a while.
I am studying for a master in computer science but I will try to pivot into cyber security because I do not wanna deal with this shit.
@@spumbibjorn Sam altman seems to be wanting to change the class World system that we have in west right now
@@spumbibjorn Heads of companies can be replaced (they cost too much). Shareholders are the end game.
Sora is sorcery in this era of AI. The leafy elephant was mind blowing to another level. 😍🙇
Yeah, I've decided to drop the creative field. I love AI, I hate that it's going to replace me, but I can pivot to a new field and continue my art as a passion. This is just how it is and I accept it.
That's a survivors mentality
I would recommend you keep track of the copyright lawsuits directed at AI companies right now.
@@mktwos Which will go nowhere. The Genie is out of the bottle.
@@dirremoire we will see.
ONE OF MY ALL TIME FAVORITE CHANNELS AND VOICES AND MINDS --- YOU SIR HAVE MADE THIS WORLD BETTER AND WE THANK YOU
2:30 as a tour guide of Niagara falls I get to see the falls almost every day, and this isn't even close to what the falls looks like, the horseshoe falls is 2,700ft across, and the American falls is 850ft across, this looks like neither
How's your hearing? That must be a major concern.
You must be fun at parties.
No one really cares bud
Why are people giving negative replies, this is a useful comment for people who haven't seen Niagara falls for a long time or even ever...
@@The9thDoctor She's underlining the value of real shots of real things.
Legit... are humans doomed? What are we all going to do for work? Signed a scared designer.
farming
AI still can only replicate - so at the moment it can't really be creative or stupid - so humanity still has 2 things going :)
@@OmguserrCreativity is often just combining known things in a new way. Just like what Sora is doing.
@@Lerppunen often ...
AI is already being used to create all new new alloys, polymers, drugs etc. Anything you can and can't imagine.
You don't need to be human to be creative. If there is an AI that imitates any form of creative sentience(learning from human creation) it's output will be creative. It will also combine that with the creations of the natural world it has also learned from.
Just watched his video! Haha perfect follow up vid. Thanks mate!
Congrats, your 'sponsor message' is the first advertisement I ever considered responding to.
See uou guys in ten years when you can upload a book and generate an entire movie
Make that two years at the current rate...
I would say in less than 12 months it will be near impossible to distinguish between AI & other sources.
The line between truth and fake is gone, is it a good thing?
No
These Sora videos for some reason always feel very similar to what I see when I dream. Cant really explain it but they feel close.
They definitely trained this on game engines and 3D programs, and who knows what else. I'm very hopeful open source will get to this stage in the next year or two, because I'm dying to pop the hood on something like this locally. Running things locally and having full control is really where it's at for me.
It's so fun looking at really good Sora videos and trying to spot the errors,
6:38 I like the two holding hands that walk off a cliff lol. Not to mention the floating people beyond them.
Yeah we should start early on becasue it gonna get progresively more difficult
Good stuff. Thanks for the video.
It's really the pace of progress that's cool. AI seems to have the most problems it seems with hands, feet, and tires interacting with other matter.
p.s. that is the smoothest dirt Rally track I have ever seen a car race on!
What a time to be alive!
What a time to be alive 😂! Love that!!
Damn! and in so little time, it has reached interesting places.
Exponentially more compute for constantly diminishing returns isn't really my idea of improvement, but okay. If I could see *one single video* at any resolution or length, which was good enough for any commercial application, I'd be completely convinced this was going somewhere. As it is, it seems like every iteration just gives us more and more refined unusable nonsense. Nothing is improving in ways that matter.
Amen
Imagine the opportunity cost of having the brightest minds in the world reinventing the wheel.
The 'opportunity cost' of never re-inventing the wheel, would be failing to have invented the wheel.
What do you mean?
@@cosmiclounge Well, I can’t argue with that logic. In any case, we already have plenty of videos and people can easily make more with current technology. There is a limited number of highly technical people. They could be working on the important problems that are facing our species and the planet. For example: disease, crop yields, pollution, energy, transportation, etc., but instead we get artificial cat videos (and that’s the best thing) because investors see the best prospect of a profitable IPO in that. Our fates are tied to Silicon Valley because when you put your money in the bank, someone who doesn’t own your money sends it there.
@@daroldfuapse6178 doesn't help either that the videogame business being a multi billion dollar industry. It's like they are trying to cash out on the bubble that is the entertainment industry rather than produce anything worthwhile
@@daroldfuapse6178 The people behind this stuff aren't building this tech just to generate cute videos; video models are a byproduct of their attempt to create AGI, which they believe (correctly or incorrectly) will turn the planet into a utopia. They have the same goals as you
Game of Thrones will be fixed at last...
The elephant made of leaves was my favourite, can imagine that in a game or film foor sure ❤
Its awesome. But there are a lot of glaring issues. It's definitly not simulating reality. Rather, its more like a strange dream version.
this is the worst it will ever be
i love your work
Fantastic video ❤️
The visual concept of that robot is VERY interesting! Good material for a game.
No it's not. It's an averaged derivative of hundreds of robots I've seen before in cheap games and cheap sci fi movies.
Regarding the low res crumbled duvet @1:28, I wonder if they used something like auto generated unreal engine world for training? This would give them training data with precise annotations...
7:00 a pillar of the Golden gate Bridge is in between houses. But overall it looks cool like the San Francisco Rush arcade/n64 game
sora ai reminds me alot about my dreams i love it
Stating that "everyone will be a film director" is a bit too much. It's the same that saying that everyone is a photographer because they take pictures with their phone, or illustrator because they can use some poor GenAI. It will for sure democratize and infinitely speed up a video production, but be a "film director" requires a large set of skills that goes beyond generate animated image.
When he says his catchphrase now we can all feel that it really is as this is all starting to get so advanced at an increasingly quicker rate it's going to change the entire world over a very short period time. I go back to the Seventies and the technological advances that I saw and experienced first hand over decades that gradually changed the entire world is being repeated with AI. Except that progress is going so fast now it's mind boggling. I imagine if you jumped forward in time a decade to 2034 it will be so advanced you'd be the technological equivalent of someone travelling forward in time a hundred years from 1924 to 2024. Better strap in.
It's an AI voiceover...
None of this is fluid simulation, fluid control, or any actual understanding of physics... like all text to image/video systems, it's simply combining similar elements from the data it was trained on, which ends up being convincing when given enough data to draw from.
incorrect, it does have understanding of real world concepts, it just not right as of now, it will keep improving in the future
@@bottomtextdanny3050 no, it really does not. It doesn't have any concept of the physics of flowing water, nor the rules behind light transport. It simply knows what a scene might look like when described as having "flowing water" or "a glass table". When specifying something like "Niagara Falls but with paint" it draws upon data of the appearance of niagara falls and of what paint looks like, and it just so happens that the only real correlation between the two is the regions with water since paint is often depicted as being poured. This correlation then causes those regions to share characteristics of both subjects, without even the slightest understanding of fluid dynamics.
It's kind of difficult to tell whether it can learn or has learned physics concepts. The laws of physics are latent in the data. Or, at least in data captured from real-world footage. So, there is a possibility that these laws get encoded approximately in the neural net. In this sense, it is not doing fluid simulations intentionally but rather by consequence. There is however a lot more happening in videos that it also needs to encode, like objects, their moving components, their color patterns, and camera angles, light interactions, lens types, etc. If we are capable of extracting fluid simulation from the neural net, it is likely going to be a poor man's fluid simulation.
I’d just like to say that this is a fascinating conversation to read.
@@baerververgaert1308 You are correct. People don't get that AI is far less I than they realize. It's just a visual auto-complete based on stealing other people's hard work without consent.
Still blows my mind.
I'd love to see what they haven't released yet at openai. I hope you have a good time.
6:48 the rabbit on the left has a pair of extra limbs...t-rex paws
No, it's a bowtie, like the two middle ones
Rabbit centaur
0:19 "
...and you will see **HAARRR**"
I started learning 3d a year ago. Watching Sora AI now, I'm both amazed, scared and depressed
we will always need good artists, don't worry CEOs probably still won't be creative enough to create stunning art. We'll need artists to generate art I think. And who knows maybe "handmade 3d Art"instead of AI art will be really precious in the future haha. Seriously tho I don't think learning art will ever be a waste of time, it will make you more creative so keep it up!
I get how you feel. You will always be able to practice 3D workflows. Same as how we can still paint. Push yourself and you’ll find a niche in the industry
Problem is not we need artists ,problem is we will need less artists.
Design furniture to 3D print and sell it.
What a time to be alive!!!
plot twist, all videos are post-processed!
Always reminds me of the movie _The Thing_ when I see A.I producing what is an expected manipulated representation of what we will be fooled by.
Sora is one of the coolest things I've seen yet. Still wonky in some videos, but really impressive in others.
It's interesting that nowadays the greatest advances are driven almost exclusively by computation, not mechanical engineering or anything else. Compare it with big inventions in the last 100 years ... this time is over. And I say that as a materials science engineer.
Software eats the world.
4:24 thats interesting when i try to create cocktails that look a bit like space/stars etc with bing image creator it also appears to form honey like piles instead of beeing proper liquid
They had this internally a year ago, i can't imagine how good the text to vid model they have not shown yet is.
Though we also know the limitations of ai image generation already and what it can’t do because it emulate and mix visual patterns but not understand why. Totally different from reasoning in text.
I love how the car has lego wheels x)
What a horrifying time to be alive
I love these narrating robots
It's weird to think that most people will listen to your voice and think it's AI generated. What a time to be alive!
It’s probably edited and composed together if it’s a longer talk and that’s maybe not always natural
@@hombacom idk man, nothing about his voice sounds natural... It is, but it's hard to accept. It sounds like a rare accent combination that resulted in this abomination. Of course, that's mean to say but my god, I can't help myself. It sounds to unbelievably fake. I wish him well though, but i can't tolerate listening to him for more than a minute, and this is a 2 minute talk, lol...
It is- Actually- 100% AI generated- Or sampled- AAND- you can tell- because- It repeats things- AAAND-
@@Gerlaffy Hmm, but his channel has been around longer than AI has been good at voice imitation I think.
@@Gerlaffy Well, I just viewed a video he uploaded 8 years ago and he sounds entirely different and natural sounding. I think you're right.
I've noted an effect in less carefully done computer art before, that if you give something a passing glance, everything seems okay, but if you study the image, or replay the video, you begin to note inconsistencies with reality. Often it just takes a more careful artist to catch them and correct them. In the case of AI, it will take a more careful artificial artist, or otherwise be left to humans to discern and prompt correction.
1:02 the plant has leaves but we now know it's soul doesn't .. 😄
and its so consistent too hahaha
I feel sorry for the adult industry, animators, graphic designers, 3D special FX'ers because SORA and a.i will take over their jobs.
boowomp
It will take over some peoples jobb for sure. But it will take a good more while before we see pure ai driven companies. Most people see it more like a incredible tool to aid artists in creating.
I gotta go to school but I wanna watch this
They have gold mines in form of human created content to train their tools on. Image the crap we will get out of tools that train on other AI created content available online which will be soon more than human created content in very near future.
They should somehow preserve human created content for future context or we may soon have very superior computers and powerful processors but very crappy ouput from these AI tools.
Okay sora generate the entirety of Better Call Saul season 2
Nice. Looking forward to future progress. You know it’s getting good when the comments are full of designers whining.
No, it doesn't use polygons.
The training data? Was what he was referring to?
He wasn't saying they did, but rather might be emulating distanced based LOD in engines that use polygons.
What car was suposed to be?
A late 2000 Suzuki Swift?
A Honda Civic EG6?
A Hawkeye Subaru Impreza with cut back?
A Renault Clio Sport?
What a confusing amalgamation...
car
It's the platonic concept of a car lol
I can see there will come a time where you'll have some kind of software where you can go into different sections for making a movie!
For example:
You first prompt a script for a movie that you'd like to see then in no time at all, there it is.
Then it generates the movie based on the script.
You go on to creating the speech for the movie although that could be part of the section above, maybe the A.I giving you a choice of voices that you want to use.
After that, should you need to, social effects can be added.
Then of course any sound affects.
And last but not least, creating the score for the movie that reflects the tone of each scene.
These could be done in one program, which would be great but more likely than not, individually because someone will want to make more money that way.
One things for sure, the future looks rather interesting.
What I'm curious about is when AI is good enough to make edits to things. How will discussion about a show/movie look if people can use AI to edit and continue it? A popular show comes out and someone asks the AI to edit a character out and continue the plot as if they aren't there... How will that go? Are we going to start seeing threads labeled: [BASE STORY ONLY]? Will we see people asking for the best fan edits?
I don't think limitless AI creation will stop people from watching other people's shows. We've very social and a lot of people watch something just to see what the fuss is about. I'm very eager to see what people make, and I think the best ideas are the things you never thought of... But the landscape might get weird.
@IcyLucario That's definitely very true and food for thought!
I've just been watching some things on here that show you can already create some amazing things given a little time and imagination while also using A.I.
I feel that in some ways, A.i will have a specific style to it.
While you can create many images and videos in different styles, there's still something that gives you that sense that it's A.I and that's without any failings the A.i images used to have which are greatly reduced now.
It will be interesting to see what directions it takes.
I also agree people will still want to watch real people and movies and there is some visceral about them, a gritty feeling.
CGI as well as A.I still have that, everything is so perfect feel about them.
In the real world there is dust, smoke etc which all have an effect on the world around us.
@Faizan29353 Yeah, mostly although I would also add that things seem shiny, they lack something the real world has.
I'm not sure if it has this, but it needs some kind of on the fly negative feedback loop. So it learns not to make these obvious mistakes.
ok i can honestly say those rabbits eating got me. my brain saw that as completely real. thats scary. all the other ones triggered that uncanny valley effect, but the rabbits didnt at all. soon we wont know whats real
I Wonder if you could use nerf together with sora to generate 3D environments
So what does this mean for light transfer researchers if it can easily do what was nearly impossible for you?
The thing is that what he says is not true. These kind of AI are good at generating things that LOOKS LIKE the data they learnt with, but they are not actually producing outputs based on the laws of physics. For example, if I ask an AI "Draw me the lenses of a refracting telescope" it will output something that looks like a set telescope lenses, but if you try to actually build the telescope based on those lenses it will be complete useless garbage.
Same thing concerning fluid dynamics : you can generate content that looks like wind tunnel tests of a plane wing (for example) BUT it will be totally useless to determine stall angle or generated drag.
It's not simply that "well maybe it's just a lack of computation power", the thing is that the model did not learnt light transport equations or Navier-Stokes equations, it learnt how to generate content that looks like the related physical phenomenon.
However you can use AI to generate data to help light transportation or fluid dynamics or whatever simulations, for example in the case of fluid dynamics, the small scale turbulence exhibit a self-similar behavior with specific statistical properties, which can be learnt by an AI to generate "content" in a real fluid simulation (i.e, solving real laws of physics)
@@76EliamExactly! I don't understand the need to imply that it understands physics, other than marketing.
Do people understand physics if they remember how a waterfall looks and behaves?
This is the most impressive AI this year.
What AI video platforms are available now? Are there any that allow you to add your own imagery as source material?
If you have a chance to run prompts, I'd love to hear what results it can achieve when prompted with stereoscopic 3d video. Is it possible to replicate it with any reliability?
When back then i had to use cebas thinking particles and pay for expensive software, plugins , and now things are easier, and cost less.
Ujung2 Ai yg bermanfaat apabila Ai menasihati manusia agar jgn sombong, dendam, marah2. Berakhlak baik.
just a quick question because of 1:24 , the way you speak is rising a question in my mind, does ai RENDER the 3d environment with a mesh or it just replicates three dimentional looking things because it was trained on real 3d photos? and if its not rendered with a mesh, then why dont we add a mesh to the background to give it more structure? or even more data to an image?
its sort of baking a hollow cake vs a filled one... it might look the same, from the outside but it isnt...
it definitely just replicates 3d things
Replicates in this case. Although there are 3D AI generators which you could render the scene from
Fortunately, right now it is too complicated for AI to make meshes and such. However AI is getting better at mimicking depth perception, which is what you are seeing
@@jlopez4889 this is not true though, AI models exist for mesh generation
I work in VFX and I would love to do a matchmove (extract camera movement and scene geometry) from one of those shots. There are incorrect parallax transformations already visible. Still I wonder what the actual extracted 3d data would look like.
Yes or model and sim some scene in 3D and be able to pour all the time into the important parts and just have the far mid ground and background be generated in order to save time. this means no extra time wasted on simulation for basic oceans and having to spend time on a system for scattering trees or rocks. no wasted time setting up complex texturing systems for mountains or hue differences on trees. In a way we can use this for the boring stuff and hyper focus control onto the main subject. You can also do what you said. Use sora for concept footage to figure out the shot. maybe even extract nerf of photogrammetry as position guides to get started modelling.
@@jinxxpwnage "No wasted time" in the real world = no paid salary hours and staff cuts. Thank goodness this stuff sucks and looks as awful as it does.
@@robertruffo2134 Well i'm all for tech if it's used to augment. I wasn't suggesting it's good to layoff staff. Of course it doesn't matter in the end things will fall into place for the better or worse.
@@jinxxpwnage How do you know that? History is not consistent that way. The middle class is already hollowed out. And... It's not reasonable to be all for tech. Nuclear weapons are tech AR 15s are tech, not all tech is good.
@robertruffo2134 i mentioned i was all for tech if it augments. I'll add positively as a clarification. We can't really do anything about it. Again time will tell
By the way i think you didn't finish reading my comment. I said things will play out for better or worse. I never stated it'd play out for the better.
SORA is amazing
very interesting! This is amazing. Where this will take us, I cannot entirely imagine. But it'll change a lot of jobs.
incredible tech
Ai world is going to be so crazy!!!
This dude’s dead consciousness reanimated via AI in a live audio video format will probably be telling us about a new incredible AI technology that can kill the user, but, hold on to your papers fellow scholars, because it can do this without the user even knowing! What a time to be alive!!!
This makes me feel like we are so close to having Ai build apps specific to the user's request. There is a very specific app I would like to build but don't know enough about coding to put it all together but it will change the health care industry.
With it being April Fools' Day tomorrow, I was thinking that in the future kids might be like 'Let's use AI to make a hyper realistic news report of an alien invasion and have our friend watch it' 💀
The bed quilt looks perfectly good to me. Were you expecting a sheet like you said rather than a quilt? (Doona for Aussies)
I can't wait for the day game development is just typing very specific prompts into an AI.
It has not launched but I think I already grasped one of its clear limitations (to be fair, I was thinking about it since I've heard about the way this model works). If your video is slow enough the model will start to hallucinate things like other video models because the history context would be lost (this happened in the house video for example, where the camera rotating almost feels like a 4D video, there's just no space for that bed to appear).
So in next 5 to 7 years ..we dont know what is real or made by ai..?
@@piteshbhanushali1140
We don't?