Portal combat
OpenAI, the maker of ChatGPT, yesterday announced new generative video model Sora, capable of transforming short text descriptions into intricately detailed, high-definition clips — just weeks after Google unveiled Lumiere, a very similar text-to-video AI product of its own.
Sora, taken from the Japanese for "sky", will generate 60-second films of anything from snowy cityscapes to alpine dog-podcasters using short, snappy prompts. Like Lumiere, OpenAI's latest creation uses space-time diffusion tech, but combines it with the transformer network used to train GPT-4.
While the quality and length of Sora’s sample videos seemingly supersede that of Lumiere and other rivals, neither OpenAI nor Google have provided much information about launch dates or what a typical output might look like — but, at the current pace of advances in AI technology, it’s easy to imagine big strides on both sides pretty soon.
Video manipulation
From today, OpenAI is ‘red teaming’ Sora — assessing critical risk areas for susceptibility to misinformation, bias, and hateful, violent, or sexual content — prior to public release, opening access to a few researchers, visual artists, and video creators for testing.
As OpenAI grapples with its new product's potential for misuse, fears surrounding AI continue to spread: a recent report from the World Economic Forum found that 53% of respondents considered AI-generated misinformation and disinformation among the threats most likely to present a global material crisis in 2024, second only to extreme weather (66%). And, with 2024 set to be a record election year, AI’s ability to create hyper-realistic falsified footage may well open Pandora’s ballot box on a global scale.