A litÂtle over four years ago, we feaÂtured here on Open CulÂture a set of realÂisÂtic images of peoÂple who don’t actuÂalÂly exist. They were, as we would now assume, wholÂly genÂerÂatÂed by an artiÂfiÂcial-intelÂliÂgence sysÂtem, but back in 2018, there were still those who doubtÂed that such a thing could be done withÂout furtive human interÂvenÂtion. Now, after the release of tools like OpeÂnAI’s ChatÂGÂPT and DALL‑E, few such doubters remain. In recent weeks, anothÂer OpeÂnAI prodÂuct has caused quite a stir despite havÂing yet to be propÂerÂly released: Sora, which can use text prompts to creÂate not just replies in kind or still images, but minute-long video clips.
“This is simulÂtaÂneÂousÂly realÂly impresÂsive and realÂly frightÂenÂing,” says YoutuÂber MarÂques BrownÂlee in his introÂducÂtion to Sora above. He examÂines some of the demo videos released so far by OpeÂnAI, highÂlightÂing both their strengths and weakÂnessÂes.
It would be difÂfiÂcult not to feel at least a litÂtle astonÂishÂment at the result Sora has proÂduced from the folÂlowÂing prompt: “A stylÂish woman walks down a Tokyo street filled with warm glowÂing neon and aniÂmatÂed city sigÂnage. She wears a black leather jackÂet, a long red dress, and black boots, and carÂries a black purse. She wears sunÂglassÂes and red lipÂstick. She walks conÂfiÂdentÂly and casuÂalÂly. The street is damp and reflecÂtive, creÂatÂing a mirÂror effect of the colÂorÂful lights. Many pedesÂtriÂans walk about.”
There’s someÂthing Blade RunÂner going on here, in more sensÂes than one. The not-quite-human qualÂiÂties about this “footage” do stand out on closÂer inspecÂtion, and in any case make the whole thing feel, as BownÂlee puts it, “a litÂtle bit… off.” But as he also emphaÂsizes, repeatÂedÂly, it was just a year ago that the bizarre AI-genÂerÂatÂed Will Smith eatÂing spaghetÂti made the social-media rounds as a repÂreÂsenÂtaÂtion of the state of the art. The underÂlyÂing techÂnolÂoÂgy has clearÂly come a long, long way since then, and though the mateÂrÂiÂal so far released by OpeÂnAI may feel faintÂly awkÂward and “video-gamey,” they clearÂly show Sora’s capaÂbilÂiÂty to creÂate videos plauÂsiÂble at first and even secÂond glance.
This may spell trouÂble not just for those curÂrentÂly in the stock-footage busiÂness, but also for those who hapÂpen to believe everyÂthing they watch. BrownÂlee calls the impliÂcaÂtions “insaneÂly sketchy durÂing an elecÂtion year in the US,” but he may take some comÂfort in the fact that Sora is not, at the moment, availÂable to the genÂerÂal pubÂlic. There are also explainÂers, like the one from the Wall Street JourÂnal video above, in which AI-indusÂtry proÂfesÂsionÂal Stephen MessÂer points out the tellÂtale glitchÂes of AI-genÂerÂatÂed video, many of which have to do with the finÂer details of physics and anatoÂmy. And if you find yourÂself payÂing unusuÂal attenÂtion to the conÂsisÂtenÂcy of the numÂber digÂits on MessÂner’s hands, just tell yourÂself that this is how it feels to live in the future.
RelatÂed conÂtent:
A New Course TeachÂes You How to Tap the PowÂers of ChatÂGÂPT and Put It to Work for You
DALL‑E, the New AI Art GenÂerÂaÂtor, Is Now Open for EveryÂone to Use
Based in Seoul, ColÂin Marshall writes and broadÂcasts on cities, lanÂguage, and culÂture. His projects include the SubÂstack newsletÂter Books on Cities, the book The StateÂless City: a Walk through 21st-CenÂtuÂry Los AngeÂles and the video series The City in CinÂeÂma. FolÂlow him on TwitÂter at @colinmarshall or on FaceÂbook.
