OpenAI, recognized for its widely used ChatGPT, is now venturing into video generation with the introduction of its latest model, Sora. Operating in a manner akin to its image-generation tool, DALL-E, Sora crafts high-definition video clips from textual descriptions or static images. The model also has the capability to extend existing videos or fill in missing frames. However, the expansion of video-generation AI raises concerns about misinformation, especially in the context of upcoming political elections.
OpenAI aims to challenge other video-generation AI tools from notable companies such as Meta and Google. While the creative prospects offered by these advancements are enticing, the growing concerns surrounding misinformation are equally significant. In response to this, OpenAI is in the process of developing a “detection classifier” to identify Sora-generated content. The company also intends to incorporate metadata to help distinguish AI-generated videos.
Despite OpenAI’s innovative strides in AI technology, public access to Sora has been limited to a select group of testers. However, the company has provided a glimpse of Sora’s capabilities through a small number of sample clips on its website. Like ChatGPT, Sora is built upon the Transformer architecture, a model initially introduced by Google researchers in a 2017 paper.
OpenAI views multimodality — the fusion of text, image, and video generation — as pivotal for expanding the capabilities of AI models. By integrating these diverse modes of communication, OpenAI aims to deliver a more potent suite of AI models that can better understand and simulate real-world scenarios.
To tackle the challenge of misinformation, OpenAI’s ongoing efforts to create a more discerning AI, alongside the use of metadata to identify AI-generated content, are steps in the right direction. However, as video-generation AI evolves, the balance between creative possibilities and the risks of misinformation will continue to be a critical consideration.
