PaperHub
Thu Feb 15 2024
NYT

OpenAI Reveals Cutting-Edge A.I. Capable of Crafting Jaw-Dropping Videos

A ground-breaking technology was introduced by New York-based start-up Runway AI in April, enabling individuals to generate videos with scenarios like a cow celebrating at a birthday party or a dog engaging in a conversation on a smartphone, by simply inputting a sentence into a computer box. Despite the initial videos being four seconds long and exhibiting blurriness, choppiness, distortion, and disturbance, they hinted at the future potential of artificial intelligence in creating increasingly realistic videos. Fast forward 10 months later, San Francisco's OpenAI has launched a comparable system that produces videos resembling scenes plucked from Hollywood productions. In just minutes, this system crafts short videos featuring woolly mammoths traversing snowy meadows, monsters mesmerized by melting candles, and dynamic Tokyo street views captured as if filmed by a camera gliding through the city. OpenAI is amongst several companies vying to enhance instant video generators, alongside rivals like Runway and tech giants such as Google and Meta (owner of Facebook and Instagram). This advancement could streamline the workflow for seasoned filmmakers while possibly supplanting less experienced digital artists altogether. However, there are concerns about its potential misuse for disseminating online disinformation, further complicating the task of discerning reality on the internet. Named Sora after the Japanese term for sky to symbolize boundless creativity, OpenAI's new system is currently undergoing rigorous evaluation before public release to comprehend potential risks fully. While sharing with select academics for review and feedback, OpenAI aims to proactively address any misuse possibilities through collaborative scrutiny. Operating on generative A.I., where it can instantly generate text, images, and sounds based on analyzing digital data like videos and associated captions. However, specific details regarding the training data sources remain undisclosed due to competitive advantages sought by OpenAI amid past legal disputes over copyrighted material usage. Despite producing remarkable visuals like a vividly rendered coral reef teeming with colorful marine life in response to descriptions provided by users, imperfections like illogical imagery are still prevalent in Sora-generated content. As generative A.I. technologies advance rapidly, concerns arise over their impact on identifying misinformation online and disrupting opportunities for traditional digital artists struggling against these automated systems' proficiency in image creation.