[AI Hanip News] Beyond Shock to Fear... 'Sora' Claims to Have Seen the Possibility of AGI

OpenAI Sora, Creates High-Quality 1-Minute Videos from Text
Understands How the World Works and Generates Videos
"AGI Development Milestone"... Will It Accelerate Implementation?

The one-minute video recently released by OpenAI is neither actually filmed nor computer-generated graphics. It is a video created by generative artificial intelligence (AI) using only a 3-4 line prompt (command). This AI model named 'Sora' has attracted intense interest in the tech industry. Although it is not the first AI model to create videos, and on the same day Google unveiled 'Gemini 1.5 Pro,' which summarizes an hour-long video in one go, all the spotlight was on Sora. So, why is the whole world so enthusiastic about Sora?

[Image source=AFP Yonhap News]

The 'Final Boss' of Video Generation Models Has Arrived

First of all, the industry regards it as the 'final boss' of video generation models. The length of the videos it can create is about one minute, which is quite long. Google's 'Imagen Video,' developed based on LaMDA, was only 5 seconds long, and the model from the U.S. startup Runway, considered a 'game changer' in this field, was about 18 seconds.

The quality is also high. Other models produced videos where still images moved slightly, but Sora can dynamically change camera angles moment by moment. The movements are natural, making it look close to live-action footage. Other models showed 'flicker' effects?interruptions appearing as blinking?due to stitching multiple images together, but Sora has no such issues.

In the field of video generation AI, reactions range from shock to fear. Since the emergence of generative AI, attempts to apply it to video have been in their infancy, but suddenly a runner has appeared. Lee Geon-chang, CEO of AI startup Inshorts, said, "Excluding infrastructure costs, it could replace all existing video generation models and services." There are also forecasts that it will shake the entire video-related industries such as movies and games. It is now possible to create a movie or game with just text or build a metaverse (an extended virtual world). In other words, a 'World Generator' has emerged.

Understanding How the World Works... "A Milestone in AGI Development"

The reason Sora is so outstanding is its exceptional 'comprehension.' It understands not only the content requested by the user in text but also how that content operates in the real world. Even without detailed prompts, it naturally flows the video based on physical understanding.

For example, if you input the prompt "walking on a road after rain," the only background description is "after rain," but Sora expresses it naturally like live-action footage. This is because it understands physical phenomena such as rainwater pooling in lower areas or objects reflecting in the water and expresses them accordingly.

In other words, it means it can think like a human. When we think or learn something, we do not rely solely on text. We imagine images and naturally learn laws such as objects falling to the ground when thrown by observing movements in reality. AI is the same. Initially, it learns only from text, but as it expands its categories to images and videos, it understands the world better. Jeon Chan-seok, CEO of Pion Corporation, which creates advertising videos and images with AI, explained, "When linking information in text, images, and videos, the level of thinking becomes similar to that of humans."

Experts see Sora as a sign that Artificial General Intelligence (AGI), known as the dream technology, is approaching. AGI is AI capable of general-purpose thinking like humans. An industry insider analyzed, "Sora is, in a way, a byproduct of research results, and OpenAI's ultimate goal is a universal simulator of the physical world, that is, AGI." OpenAI also did not hide its ambition through its blog, stating, "Sora will be an important milestone in developing AGI."