OpenAI’s Sora: How to Spot AI-Generated Videos | WSJ

OpenAI just revealed Sora - an AI video generator that creates hyper-realistic scenes and animated worlds in moments. But the tech isn’t perfect. There are also concerns over misinformation, with OpenAI saying it’s taking actions to get ready for the 2024 presidential election, including prohibiting the use of its tools for political campaigning.
WSJ dives into how to spot an AI video, and the ways this text-to-video tool could shape the future of content creation.
Chapters:
0:00 New AI video tool
1:00 Spotting an AI video
2:28 Landscape and animation AI videos
5:08 Misinformation and industry concerns
News Explainers
Some days the high-speed news cycle can bring more questions than answers. WSJ’s news explainers break down the day's biggest stories into bite-size pieces to help you make sense of the news.
#AI #Sora #WSJ

[@TomNook.]

"sadly, one of the horses melts into the ground" is the greatest line ever

[@jamesmcinnis208]

I thought so too. Can we take a moment to grieve for the horse?

[@dzcav3]

No pixels were harmed in the making of that video.

[@wildfire9280]

The glue barons got him 😢

[@jessicabing6222]

FALLLLCORRRRRE!😢

[@doodledogdiary3644]

​@@wildfire9280NOOOOOO 😢

[@Rvj.12]

Big tech companies should make it compulsory to add water mark “AI generated” in every picture and video generated by AI

[@ViceZone]

That would ruin the whole purpose of these tools

[@Rvj.12]

@@ViceZone plz elaborate?

[@justanotheryoutubefan8070]

Agreed

[@TheJmac82]

You mean like forensic fingerprinting that sites put on their videos? Guess what piracy is at an all time high and they remove those automatically now. There will be an open source AI tool that will remove them and rerender them. DVD's had DRM on them... took years to design, 24 hours to bypass. Blueray tried it, CDs tried it, none were successful.

[@_tjoeaon]

100% agreed, people will of course find ways to get around this as people will train their own models based on this technology. But I still think it should be enforced in some kind of way

[@0hmyl0rdd_]

With how social media runs nowadays with divided attention spans and quick gratification through scrolling, these minor details are unfortunately not gonna be as noticeable.

[@joebidenofficialpotus]

And video compression

[@chiragarjun3270]

@@joebidenofficialpotus and AI can learn and u will not see this in future and that's real scary😁

[@chillfluencer]

It's evolution, babeee!

[@davidpaul8039]

These flaws will be fixed in no time

[@tomaccino]

That's Tiktok generation for you 😂 "bUt iTs tRuE CuZ iTs ai"

[@VladiSSius]

The title should be : "How to Spot AI-Generated Videos, For Now." :))

[@TurbocrafterG]

because ai tools get better and better faster. its kinda scary

[@SwiftDustStorm]

Agreed. Actually these flaws are talked about in the whitepaper, and open AI has been pretty consistent in improving upon self-recognised flaws.

[@N0bodyn01]

Guarantee it’s already possible. But they don’t want people to know that it is because then people won’t question the propaganda that this is literally inevitably going to be used for.

[@RayZacario]

Exactly my thoughts, soon we would not be able to tell the difference; they are perfecting this technology at a rapid rate.

[@jamfilledjars]

I’m worried about AI art because I want to be a character artist. But, I’m even more terrified of the repercussions that AI videos and audio will have. Misinformation, slander, propaganda. Honestly, if it comes down to the wire, I could easily see this starting conflicts.

[@user-sf9gs2pg1b]

I’m particularly worried about how this will completely ruin any and all video evidence since it could be AI generated. I was a juror before and if the videos they showed to us could’ve been fabricated, I don’t think I could ever rightfully call anyone guilty. A large part of the evidence was video and a witness. Now, that witness could be a liar and the video could be completely fake. Then again, they had to ask for the footage and there was a sign out sheet with the guy’s name, but the other video was definitely able to be faked. Or worse, this could be used to frame someone, resulting in them getting the death penalty or imprisoned.

[@Ninten110dogs]

This^

[@myxobe]

be a prompt engineer instead

[@monkephrog5984]

​@@myxobefound the dork

[@brysoncherry9884]

I mean people have already been deepfaked into pxxn and lost jobs because of it.

[@simonphoenix3789]

yeah this information is good for about five days before the AI improves and it becomes even more indistinguishable. It wasn't even a year ago where AI videos looked like horrific nightmares brought to life. Now they are almost impossible to spot without careful observation in most cases. We really opened pandora's box here with this stuff.

[@javanpohl3036]

For sure. Even with all of the advances in CGI technology "physics" (which is a sticking point of his in this video) is something that still plagues modern animators. I have little doubt that AI will soon be able to do this much better than your average (or maybe "any") animator.

[@pvanukoff]

Right. We all need to keep in mind that this is the WORST this technology will ever be. It will only get better from here.

[@timothymcguire5126]

Fortunately, thing like digital signatures exist. I imagine in the future, camera manufacturers will build in some kind of cryptographic digital signatures on a hardware level so that it can be verified whether a digital photo has been altered by any means. Also, we don't know how far this tech will go. Making these AI models bigger doesn't always make them better.

[@Sanzarc]

I think that is what most people don't want to admit that progress will be exponential. If they admitted that then the public would react negatively to AI (even more so than now). Although, I feel like we're running towards a "Don't Look Up" scenario where the general public will understand the impact of AI in the last second while experts and those who had money to prepare will know what to expect months/years before the larger impact happens.

[@TheRadioAteMyTV]

Did we open Pandora's box or did AI do it and we just thought it was Pandora?

[@issiewizzie]

If people can't be bothered to validate, News, or sources of news, or a fake photo, what makes you think they will be willing to spend time to detect a fake video?

[@Parlof96]

Just as with everything, those who care will care and will continue to reap the benefits of living in the real world and pursuing truth. Those who don't care.....well they won't care.

[@TK-gd9td]

thats why companies are popping up in existence to make AI that detects AI lol.

[@AAAAAA-tj1nq]

​@@TK-gd9tdmajority of the people isn't going to check the video or picture with ai checker

[@thatwasinteresting3319]

It will become a problem as who will you believe. Let’s say a person videos a famous politician doing something wrong. That powerful politician screams “It’s a deep fake!” If he can get enough “experts” to agree with him then he has contained his problem. Same with the opposite too a fake video becomes legitimate with enough “experts” declaring it so. It will be an ongoing war of truth versus fiction and the general public will never know who to believe. And that is what the powers that be want. He who controls the narrative controls the people

[@SeeTheWholeTruth]

IT will get good enough, that they need an AI to verify against an AI. Then the real hilarity ensues.

[@mk1st]

Sure, this has a potential to unleash peoples’ imaginations to fantastic levels but we’ll also lose the wonder of “how did they make that?” For instance that scene in the Matrix when the camera(s) do that swing around Neo as he dodges the bullets flying overhead. The sheer effort and ingenuity to pull that off was part of what made it amazing. That sense of wonder and amazement will be gone.

[@johanngambolputty4117]

There are many people whose imagination should never be unleashed…

[@AlFirous]

From Idea straight to Final piece, there’s no process from human side. Things that magical become mundane because how easy to make.

[@phazerxp339]

I actually want to cry over this comment...

[@strvmpet]

nono.. this isnt unleashing people's animation this is ai doing the whole job

[@Doctaphil64]

The most alarming fact is that this is the early version of the tooling. What will be available in five years will be indistinguishable from real.

[@karlkelly9005]

Exactly 💯

[@iluminumfalcon8619]

Oh they’ll have a heyday with Holodeck once Holograms are invented better.

[@fg009letyrds8]

*a year

[@strangelaw6384]

so what, though?

[@alexm2889]

​@@strangelaw6384so every single thing you see on a screen will be falsifiable. You will never know if you are watching is real or fake. Imagine what that does to a world built with systems based on trust and a shared reality. Social media was just the tip of the iceberg...

[@arzalalbuchari7095]

i'm sorry if i'm exaggerating, but everything related to ai advancement nowadays has been worrying, and i'm not typically a conservatist when it comes to technology, but i believe this tool should be moderated heavily and not be available to everyone/commercialized to the public.

[@teflonda5655]

What do you think will happen when only large companies and governments are allowed to use AI? I don’t see that being a good thing either, and the option of just not allowing the public to have it is not even an option that’s viable anymore. You would’ve had to have done that before AI got to this point, which before hand no one thought it would, so soon, or as quick as it did, which means no one would’ve implemented these restrictions because we aren’t exactly proactive in most areas.

[@ketaminepoptarts]

​@@teflonda5655 yeah we should just outright ban ai generated images and videos

[@user-sf9gs2pg1b]

If it’s not in the hands of the public, it will be used by companies or other large entities and we will never catch up to them. I’m so glad that this stuff is publicly available, I just won’t trust accounts who pop up with random videos.

[@b14922]

No, I am going to use this to spread so much misinformation and disinformation

[@chrisf4268]

Right only the rich and powerful and governments should be able to use it to create propaganda and manipulate reality. That surely will make all our lives better. 😂😂😂🤣

[@TexanWarEagle]

I can’t wait to see SORA mimic realtime CCTV footage. That’ll be a hoot.

[@SolaceEasy]

Guilty! Video proof.

[@cat_quitesuit]

A good example of the word 'abusive' 😮

[@user-lx4zq5xf1t]

who said you can't? :)

[@OgdenM]

If SORA can do this stuff it for sure can do CCTV because there is plenty of CCTV footage online.

[@YoNeener]

Lol your brain 🧠 🥰

[@Nainara32]

Video footage is currently accepted pretty much as incontrovertible evidence that something has happened in everything from the prime time news to legal court cases. When AI generated content becomes readily accessible, that trust will need to be reexamined.

[@MariaMartinez-researcher]

The source of the video is the key. It is not the same a weird/damming statement by a politician made in a public venue, with multiple witnesses, filmed by a living cameraman from an actual news medium or a living witness with the original file still in the phone's memory - than a short uploaded directly to social networks by an anonymous user who opened the account the previous day. Technical issues can and will be solved. Real life will still exist.

[@ThurstanHethorn]

Is it though? I would be fairly certain saying that at least cursory glance of authenticity of videos is done before being accepted in court. Camera makers are looking to implement methods of providing authentication, perhaps an encryption to a particular device that is encoded into the photo or video. Not saying it too won’t have faults. Sure the court of public opinion is another issue though.

[@alexm2889]

​@@MariaMartinez-researcherAI just makes a number of fake videos of eye witnesses discussing what they saw. It hacks their existing accounts and posts them or creates fake accounts and creates a fake account history that can't be verified. It really is limitless. All of these capabilities are not far away.

[@quelquel5079]

AI will help to understand and prove.

[@dibbidydoo4318]

no it isn't, legal court cases use chain of custody.

[@Itsmetahh]

Plot twist : the video we just watched is automated Ai video by sora 💀

[@OxzowachiAlt]

💀

[@brAvOtIsm07]

💀

[@rogeranimates3357]

☠️

[@ECHOOZzs]

💀

[@tartaglia8579]

💀

[@MrRapmaster19]

This is why moderation and awareness are SO important when it comes to this. Some of this stuff seems obvious, but eventually, as these tools get better and better, they won't be. Not only keeping people aware of what's AI vs what isn't, but punishing those who attempt to use AI for nefarious purposes (revenge porn, fake news, etc.) with the cruelest of strokes.

[@muwaten8247]

Totally agree 👍🏼...

[@winowmak3r]

The AI revolution is coming at a time when the education system in the US is in shambles. It's a perfect storm and it's not going to be pretty. People will have no way to even ask *themselves* if this stuff is not real let alone try and convince someone who's been glued to tik tok for their entire childhood and can barely read or write at anything approaching an adult level.

[@dexterwestin3747]

It won't work, it is a forgone conclusion that this will happen and how would someone punish others who use AI for "nefarious purposes" as you say? You would have to find them and they could be doing this from anywhere including Russia, Iran, North Korea, etc. We can't even keep granny from being swindled over the phone by someone 10k miles away.

[@dianapohe]

This just made me realise that AT LEAST actual revenge p0rn victims will be able to defend themselves by claiming the videos were just AI and not real

[@schmutzsteph]

This video is a little self soothing. The problems outlined with AI video (and therefore allowing us to 'not worry') are just today's Ai challenges. Each problem outlined is just a heuristic waiting to be developed. Standards and regulation via things like the Content Authenticity Initiative must really be explored by both industry and regulators. The artifacts outlined in the sample videos as 'evidence' of Ai are too easily solved and cannot allow us to be complacent.

[@otaku-chan4888]

this video never claimed to be the solution against 'tomorrow's AI'. it very clearly mentioned it's only talking about recent AI videos. There's nothing to be gained today by simply saying "yep problems and ways to spot flaws are useless because they'll all be fixed soon, it's time to give up and wait for the worst".

[@rasimbot]

AI models perform well when they are trained on human-created materials. However, as people increasingly flood the Internet with AI-generated content, these models will begin training on that AI-generated content. Theoretically, this could be similar to copying a VHS movie from a copy that is itself a copy, and so on...

[@moos5221]

Shouldn't be a problem to teach AI to detect AI videos and to ignore them.

[@paige66666]

There was a recent article basically stating this issue with written content online, the speed of creating it has been exponential so the overall volume of content is now majority AI drivel, which now makes the content valueless essentially. The arms race will continue for content vs consumption

[@kiraneu1529]

​@@moos5221 It's a huge problem because it CAN'T detect that accurately.

[@Isabelle-ox7go]

That's so true and something I haven't thought about. Hopefully that happens and people will realize how stupid and pointless AI is. What good does it even do ? It just takes the joy our of content creation...

[@simulping4371]

A.K.A generational loss

[@infini.tesimo]

It really will be the last time that this is the worst it'll ever look. In 5 years I'm absolutely horrified that this may not have legal coverage for being framed for crimes not committed.

[@dexterwestin3747]

Or for committing crimes and then saying the CCTV footage is fake and convincing a DA and jury that the video can't be trusted.

[@OgdenM]

5 years? Try 5 months.

[@boslyporshy6553]

Moneys on it flatlining. The tech is dependent on the designers.

[@alexm2889]

​@@boslyporshy6553not when you apply AI to building the next iteration of AI. The singularity is just around the corner.

[@MildMisanthropeMaybeMassive]

You also know that people will actually commit crimes and claim that the real evidence is AI generated.

[@Wojtackic]

There should be some kind of watermark or something so that there's a clear and noticable sign that it's ai generated. I don't know how easy it would be to remove tho, so it might not be that foolproof, but there needs to be an indication informing the viewer right away what's ai and what's not imo

[@DonLee1980]

"this is what AI still doesn't understand about the world." OpenAI: challenge accepted.

[@medamine3142]

openai taking notes from this video LOL

[@xxxkunti]

Get her Jade! 🧏🏿‍♀️

[@HelamanGile]

​@@pedecadonstudios714I mean it's better to have access to it than not have access to it would you rather corporations only have access to this technology?

[@user-wd8wx5md5z]

​@@pedecadonstudios714 It is like the invention of photography that made realist painter almost obsolete. The mobile phone's cameras made photographers almost obsolete (everybody can take pictures). Now we democratize the possibility for everybody to be an artist even when they have no skills. AI generative diffusion model have very basic language understanding and zero understanding of the world. We need to improve it a lot before hoping making everbody obsolete.

[@BlownMacTruck]

I don’t think you understand how current AI works. It doesn’t “learn” or “understand” anything. That requires cognitive functioning. All it does is operate within parameters that are specified by users. It’s the same way a robot vacuum cleaner has no idea what the boundaries of a room are. It eventually gets it by being given parameters (in this case by mapping its environment through trial and error) but it doesn’t “understand” what a boundary is, whether that’s a wall, a bookcase, a person, or an animal.

[@NinjacatCreative]

Its really important to know what is AI generated and what not, people can probably make a full realistic video of you commiting a crime with AI. Seeing how far its come since one year ago, its crazy. They should make something like a watermark to recognise AI from real

[@hessiboi_]

the way she delivered the horses melting, truly broke my heart 😔

[@davidthedeaf]

It is okay. Horse is in a better place now. It melted right into a little girl’s dreams and she went off into a rainbow to meet the angels.

[@hessiboi_]

@@davidthedeaf 😔RIP

[@TailspinMedia]

the problem is AI is still in it's infancy, all these shortcomings will be ironed out in the next 2 to 3 years, if not sooner.

[@SolaceEasy]

Where's my personal jet pack?

[@mountainvalleymedia5836]

6mo-1yr tops GPT-5 is already here, the drones aren't yet told, but you don't collect 7 Billion dollars trough fundraising if you just have a flipchart and a dream

[@mountainvalleymedia5836]

@@SolaceEasy Are you a member of the 'Elite'? No? Please be quiet, these toys are not for you work and pay your taxes!

[@Snowboarder54688]

Exactly.

[@nomadv7860]

@@mountainvalleymedia5836$7 trillion lol

[@tritonwe8937]

Remember the times where we would see ghost footage on youtube and say "This is photoshopped" or something like that ?

[@user-sf9gs2pg1b]

Yeah, lol

[@ACE-kt6sc]

Not gonna lie - I was half-expecting a marvel styled "this video was generated by SORA" after end credits

[@iiwi758]

The problem (or maybe the current hope for artists) is that you have to rely on the AI to get what you mean from the prompt, which likely won't be exactly what you had envisioned. Once developers figure out a way to make the AI animate based on a user sketch, and the AI keeps a consistency between prompts, then there will be a lot of people out of a job the next day.

[@WolfMoonWings]

So good for society whoooooo

[@iiwi758]

@@WolfMoonWings I didn't intend to imply it's a good thing if that's what you mean.

[@toodled1340]

@@iiwi758it's sarcasm (i hope)

[@kinyacat5919]

As someone that dreams to be an 3d animator, I fear my dreams will be useless once I turn into an adult

[@BobMarley-uz9mn]

@@WolfMoonWings No thats bad, means no income for people and millions will be on streets while the rich brag about how AI made them millions

[@paigeupdike9057]

I can never place why, but I tend to find ai vids/images unsettling and sort of gives my brain a headache (I like gaming but I get a similar feeling when I play too) I think what also unsettles me is a lot of the time the videos/pics seem empty somehow. Pieces that don’t seem to have meaning. One dimensional… but the single dimension doesn’t even seem coherent with itself. But as everyone is saying, won’t take much more time to iron out a lot of this

[@vvelvettearss]

i did too. all of these videos unsettled me and the grandma terrified me so far I can see the movement is too fluid to be real and thats what gave it away

[@bpmotion]

Let’s get back here in 6 months and see how this aged.

[@angelmauriciohernandezgarc9857]

Checked out specswriter white paper maker - holy cow can it write!

[@pastorofmuppets9346]

Man the landscape shots especially are wild, id never be able to tell

[@Phattadon02]

These AIs are inevitably going to be flawless one day, and that kinda fu*cked.

[@varunprasad9]

LOL, when they start feeling bad for the horse at 3:27

[@jamesmcinnis208]

I was sobbing uncontrollably.

[@Talia.777]

🤣🤣🤣😂😂😂

[@coolbluereview]

Why doesn’t the government require AI created content to be labeled as such?

[@justinvt]

There is really no way to “fingerprint” a file like that now, although openAI DOES do what you’re suggesting , with C2PA. The thing is, a person can still hold their phone up and record the monitor, or use an analog camera to record the monitor. For an HD video, you only have to look at a million pixels. I’m sure you can make something “hard to spoof” but when processors can evaluate the entire decade of digital output from the 1990s in one second (and I think we are way past that mark) , I certainly don’t have a great solution.

[@coolbluereview]

@@justinvt I just know that this is a recent law that was passed in China a year ago. They are very strict about the enforcement. I think the US should take AI regulation just as seriously and not let the industry police themselves like they have done with social media and the finance industry in the past. AI can impact our world in bigger ways than the internet or banks have had.

[@coolbluereview]

@@justinvt I believe they just require the platforms or companies that are involved in the creation of AI content, to watermark everything created by their software. I’m not sure what else the law does.

[@paige66666]

Because capitalism, they are afraid laws restrict business competition and innovation and would rather react than be proactive

[@coolbluereview]

@@paige66666 Perfect example of the 2008 recession.

[@007SuperSoldier]

the examples of weirdness pointed out in the AI-generated videos remind me of LSD

[@alexm2889]

Yeah good point. Some of the imprecise nature of AI really feels like how the human mind works. I'm sure whenever they actually figure out how it works, if they ever do, we'll see it's radically different.

[@Zegeebwah]

When you have access to everything nothing means anything

[@SolaceEasy]

AI generated wisdom?

[@jmason61]

Ummm.... still wondering about it but sounds Zen

[@alexm2889]

Totally true. We value a an amazing Pixar film or video game so much now. Once you can just call for one to be created on the spot, all it's value disappears.

[@alpacalover0]

​@@alexm2889 Speaking of Pixar, this AI stuff is pushing us towards a society very much like WALL-E. Instant gratification, no matter the cost.

[@marcusmoonstein242]

"It will be a long time before you can make a movie with SORA" Really? At the speed this tech is developing? Unless by 'long time' you mean six months.

[@tomlabooks3263]

Exactly. People do not understand this. Nothing that anyone can do about it, anyway.

[@paige66666]

With a little editing skill and some levels of control for prompt writing, you can basically make something a la everything everywhere at once. Will we be making narrative dramas in 6 months, no, but nonlinear video content still is very much at play

[@No-longer1]

I think you are underselling the consistencies required behind movies

[@AlFirous]

@@No-longer1Yup, and movie is not only about video. There’s more behind that.

[@mojave5661]

to me, it looks like ai image generation has hit a plateau in late 2022, and some say it's gotten worse since then (because some models are now being trained with other ai images). I originally wanted to use ai to make my hobby easier, but I've since spent weeks sitting behind Photoshop fixing all the details it missed. and my standards aren't even that high.

[@onebear6504]

Do remember that technology companies (big ones especially) always have technologies in development of 2-3 generations ahead, just not released/announced yet so the public doesn't know. And what we've seen of the video is just a pre-release recording probably recorded long ago in preparation for public unveiling. What the AI is capable of now would probably be different, better.

[@Andmau2552]

"How to spot fake videos" "Check if it feels fake or if someone has 3 hands or objects dissappear" Thanks, very helpful

[@deetrvl4life875]

Best reason ever to start planning a total disconnect from all things Cyber and begin living in the ‘real’ world again! What an adventure!! 😊

[@vvelvettearss]

@@kisamehoshigaki1492 agreed. it really has gone.mad it saddens and frightens me . even moreso that they are just calmly proceeding with this and will do so even when problems arise maybe it was always waiting to go mad but i never expected this :(

[@spacecowboy5274]

​@@kisamehoshigaki1492 "Go back to nature" So go back to millions dying from diseases and starvation? There is nothing inherently special about human creativity. I don't see any problem in it being automated.

[@coolstar7819]

yes!!

[@giotto_4503]

This is just a year of progress. A YEAR.

[@Penelope416]

But I'm sure that they've been working on this project for longer than a year, to even get to this point.

[@zac9181]

This is something that I think is required for everyone to watch. It should be shown in schools around the world!

[@michaelfreydberg4619]

One way I can tell, is I’ve seen some videos that are like “important news flash!” And if it sounds too incredible, I google it and often don’t find the info. That’s one way to tell…

[@vvelvettearss]

prettty good thinking!

[@lestranged]

These glitches are funny and easy to spot now. But Over time they will improve these red flag flaws and it will be harder and harder to spot the fakes. Which is scary. Like it's normal for tech to fix bugs and improve but this is a technology I don't want it to get TOO perfect.

[@grimnirnacht]

He didn't even mention the appearance of the random hand in the blanket? That's a little more telling to me than a third cat paw. Like is it detached and just rolling around his bed? Or did his blanket arm suddenly grow flesh

[@Zephyr-lg5ne]

Congrats you just gave open ai a step by step tutorial on how to make their videos indistinguishable from real life.

[@will-todman]

I know this is scary, but as an beginner film maker with a camera and a greenscreen, this opens so many doors in regards to special effects and sets. This is the kind of thing it should be used for. Not identity theft like it will be

[@shoberino3898]

Exactly, creative expression will explode with tools like this. And animators might even get to see some sunlight

[@KratosUmar]

SORA is in first Gen Product...let them release couples of updates.

[@user-lx4zq5xf1t]

the thing is after this tool will be released fake news will skyrocket because it will be impossible to spot the difference between fake and real, event with C2PA ..

[@absence9443]

Wow thank you so much for this tutorial!!11! I'm sure it won't be out of date in like 2 weeks

[@manzoj5898]

Appreciate this guy coming on and helping us

[@danielrus7117]

Last year we were concerned about noticing differences in Ai generated images, and now we already in the same with videos, this is going so fast! 🤯😳🤨

[@brandondoggett4538]

‘one of the horses melts into the ground’ SENT ME

[@Kleinage]

Sadly 😂

[@Talia.777]

@@Kleinage 😂🤣🤣🤣

[@lennonsickels8913]

Weird connection, but I went through a period of time with consistent night terrors, and to combat this problem I learned how to lucid dream. There are a lot of tools to figure out that you’re dreaming while within the dream - but incorrect physics, inconsistent matter and time, and a certain lack of ‘realness’ to human hands specifically are all pretty major clues. Kinda freaky, in a way, that these tools translate so well to figuring out if a video is AI generated. Curious to see if any other current/former lucid dreamers have noticed the same thing?

[@StevenAkinyemi]

5:51 Looking at how even the AI industry was caught off-guard by the release of this model, I say these "experts" are a bit arrogant and burying their heads in the sands if they think it will take a long time for that to happen.

[@huntrrams]

The AI in Sora is still in its early stages, but I'm afraid of how advanced it will look 5+ years from now.

[@pauliewalnuts240]

Yes errors like having 6 fingers are obvious, but these videos otherwise look very realistic. However, Your average human doesnt notice the inaccuracies due to physics. It also causes people to question everything, especially topics they may disagree with. People have begun criticaly examining factually real videos and finding seemingly inaccurate details that actually are realistic aspects.

[@a_921]

Yup. Also, for short videos, unless you are actively looking for "something", you will miss it. I mean, if it's just a cute / funny / random video, why would you stare at somebody's fingers to see if there are five or six? I found the most creepy thing about "grandma" her grin

[@Dev1nci]

2:50 except that waves do move counter-intuitively sometimes due to reflection and wind and on centuries old mansions staircases do end up leading to 'nothing' because areas get enclosed, also the one staircase circled leads t a doorway...

[@cbusse7842]

yes how to spot an AI video "based on the very first AI video generator there is", there isn't even a comparison. by anyone standards this is already amazing, and all you need to do to fix it is prompt it different. wait until Sora 2 comes out. wanna hear how you spot it then

[@OgdenM]

This actually isn't the first btw. It's why it's so good.

[@JJ-ctrl-alt]

It’ll be a long time before we can create movies? Naive thought. It’ll be sooner than later.

[@paige66666]

Haha exactly, "long time" in what sense? movies themselves haven't even existed for a relatively "long time" in human existence, so really how long would it be until movies are autogenerated. It's like saying because people didn't print books for milennia there's no way that the tech could drastically change the years after the invention

[@janechoy2073]

Great video; not sure how log it'll remain relevant.

[@new-bp6ix]

Artists are good at seeing this Honestly, even artists see things that ordinary people do not see it

[@AmanSingh-lv9qp]

1:14 did our Sora just took over magician career

[@shotbyoliver]

Sounds like the voiceover for this video might be AI too. Hey @WSJ, who did the voiceover?

[@AeyGee]

Get your ears checked ASAP.

[@goat7844]

@@AeyGee at least he or she questioned it. You on the other hand believe anything you hear or see.

[@nexenzy5825]

same as the trend of "look at the hands" for ai generated images, they evolved and are generating better hands. all you are doing is pointing out what the ai coders need to make better

[@lauriivey7801]

I love the half-horse walking down the street in the western town, and the guy that was riding one of them when half of it just disappeared

[@DamBevers]

The scariest aspect of this is that lawmakers, especially in the U.S., are too geriatric to perceive the danger, and so many politicians, especially Republicans, don’t run for office to make laws. They represent their donors’ special interests only and engage in useless partisan obstructionism instead of law making to regulate and mitigate the risk and dangers of Ai.

[@justanotheryoutubefan8070]

Yup. Absolutely terrifying

[@lilgrandpa9005]

"Especially republicans"? Yeah, cause there's so many young and hip democrats making laws about ai 😂 and they NEVER have big donors

[@sungjane]

The Democratic Party did not ban ASML from going to China！

[@goat7844]

💯🎯

[@RichOffKs]

I mean, these errors are only for version one

[@iluminumfalcon8619]

The dude calls him an Expert but didn’t even mention photoshop videos can be touched up fixed over time.

[@watercat1302]

AI video checker is gonna be a real job in the future…

[@anonymouse9833]

It’ll be a job for another AI though haha

[@amedeobazar9423]

This is just the beginning 😮

[@aquamanGR]

Great video, and it all makes sense, since the AI is not trained using any laws of Physics.

[@vistabadboi]

The problem is people are stupid and they believe this stuff

[@viviendomisabatico1587]

Gran informe, Christian, gracias

[@CatabolicWaffle]

The applause during the birthday clip was some body horror stuff right there

[@pixieanna1]

It just so crazy to see them gone, when they were there from day 1 of my introduction to anime.

[@maxwellvindman7212]

0:42 you can really tell the difference. even in older pixar movies they were filled with tiny details the ai doesn't do convincingly

[@cloudzero2049]

The end says it all. Were in the early stages. Soon even top professionals won't be able to see the difference.

[@saumitrachakravarty]

We learn how the world works via both induction and deduction. AIs like SORA just learns by induction. You cannot just replace deductive reasoning by feeding tons of training data through induction.

[@OgdenM]

So you say, but I say you can. Partly because there are actually people in the background training the AI.

[@katobrucelee08]

2:46 the person on the left staircase walks and fuses into the stairway wall

[@LookjaklooK]

"The stairwell goes to nothing, that's how we know it's AI" - 2:40 No, that's a perfectly human thing we do all of the time XD

[@jfess1911]

Pertinent phrase "these are the things that the AI still does not understand about the physical world". Of course, the logical extension is "... but it is still learning". It's only a matter of time.

[@lori25hu]

AI looks at this video and 🤔 I am learning ☺️ thank you for pointing that out

[@MariaMartinez-researcher]

To the people saying and thinking these tips will be useless in the short term and fake videos will lead to chaos: The source of the video is the key. It is not the same a weird/damming statement by a politician made in a public venue, with multiple witnesses, filmed by a living cameraman from an actual news medium or a living witness with the original file still in the phone's memory - than a short uploaded directly to social networks by an anonymous user who opened the account the previous day. AI videos' technical issues can and will be solved. Real life will still exist.

[@justinvt]

I totally agree. There could be a perfectly believable video of me committing an atrocity released right now. If I am at my sisters birthday party in Puerto Rico with a hundred other friends, you can’t make me “seem to be” in Japan on the same day. It’s an alibi, and the justice system relies on them pretty heavily. My phone is with me 98% of the time. I don’t have control over where my phone says that I am. But it’s very accurate

[@yellowbird5411]

With the attention span most people have these days, no one would notice a five legged cat or two hands on the same person that were very different. Give it time. It will get better. But I read a commentary that said by the very virtue that AI will be using more and more of it's own limited images to make new images, AI will implode on itself in limitations of images. I thought that was a very interesting theory.

[@Seehart]

This is excellent advice on how to detect fake videos over the next two to three months, after which these flaws will be almost nonexistent. Of course when this becomes available to the public, it will likely be hamstrung to prevent many abusive use cases, but the underlying technology to produce these videos will become ubiquitous. Uncensored open source implementations for these technologies only lag the mainstream (OpenAI, Google, etc) by a few months or so, so safety measures only have a temporary effect. Ultimately, we should only trust sources that are verified by blockchain or some other public key encryption standard. But that only verifies that the message content is intended by the verified author, not that the footage is real. None of that will matter to the 99% of viewers who don't know (or don't care) how to fact check or confirm verification.

[@_dimitri_1242]

These videos explain what being on a certain substance is like. Eeriely similar.

[@habby5687]

Thank you. Thank you so much.

[@judemorales4U]

Yeah, people with bad intentions could do serious damage.😮

[@AnthonyValli]

This is why I gave up on entering the animation industry. Glad I didn’t spend too much of my life trying to break in.

[@MarcioSilva-vf5wk]

That beach areial cliff shot it simply cannot be detected as AI made

[@proxyhacker89]

This is so scary. Imagine needing a pro to tell if a video of you committing a crime is true or not. In the meantime you're in jail being shanked by someone blatantly believing the victim.

[@elliart7432]

You can also spot them by just asking “are these people acting like real humans?” There’s a video of a grandma blowing out birthday candles… on a countertop facing away from her family as they sit in a row of chairs applauding her doing something they can’t even see

[@JL-uu3lv]

I watch shows with my toddler and some of these kids shows and movies are so real i can barely tell. Imagine growing up seeing that.

[@willow7466]

There's a lot of people who seem to believe that the ai will be able to create indistinguishable images from reality eith a few years, this is still just a prediction. We won't know how much the quality will improve until the time comes, but for now id say hands and forearm movement are dead giveaways. If you want a reliable reality check.

[@Lahiru_Sampath.]

This is only beginning

[@angelgomez-tz6jr]

This is very cool tech!

[@gnomidomi]

These tips will be relevant for 2 months

[@DamBevers]

Good reporting. Everything said here could likely be amended in a single software update in the near future. Training Sora and the likes on physics, proportions, anatomy and physiology, and edge cases (anomaly specific) would iron this out and make Ai generated content impossible to spot with the naked eye.

[@thelostcosmonaut5555]

I remember the Battlestar Galactica episode where one of the characters were framed via a video created with technology similar to AI. Creepy.

[@Arcwol]

Still pretty good, regardless. Wow, that's is crazy.

[@MariadelosAngelesChavalaVazque]

Neat video!

[@NM0047]

Nice one.

[@Seyfettin.a]

Bro, you’ve misunderstood the situation. OpenAI has already stated that Sora has some deficiencies and has said that they will address these issues soon.

[@tom_selleck]

You’re also missing the point.