Last weekend, Open AI released a new service called Sora. Sora allows you to create AI videos by entering text, and has been receiving rave reviews from reviewers since its release. The demo videos released through a few simple lines of text are as realistic as scenes planned, directed, and filmed in commercials or movies, such as slow-motion footage of a middle-aged man eating a hamburger, a low-angle shot capturing a Japanese street with two people walking, and a low-angle shot capturing a Japanese street with two people walking.

On the same day, Open AI expressed Sora as a world simulator in the technical report it released. Abstraction through text and its connection to the parameter space make this level of realistic abstraction possible, and will have a huge impact on the reasoning capabilities of future GPT models. The significance of this is enormous in terms of the influence of AI on human society.

However, Sora still has many limitations. For example, a person takes a bite out of a cookie, but the bite mark disappears afterward, and a person running on a treadmill runs with awkward arm and leg movements facing backwards, not forwards where the control panel is located. There are shortcomings in accurately understanding and showing spatial details within the user-provided prompt, changes occurring over time, and so on. These shortcomings in demonstrating cause and effect, and difficulties in implementing physics in complex scenes, are evidence that AI still does not fully understand the world. However, considering Open AI's 2023 announcement that it plans to train AI for another six years through its contract with Shutter Stock, which has 35 million high-resolution photos and videos, it is not difficult to expect that AI-generated images and videos will become even closer to the level of human-made content in the real world.

In his 2013 book "The Origins of Monsters," British archaeologist David Wengrow argues that from primitive Egypt to Mesopotamian civilization and the Mediterranean Iron Age, when cities were formed, civilizations flourished, and political and commercial networks expanded, there were no regions where visual images of fantastic and complex unreal creatures were not created and spread.

Graphic depictions of composite monsters, such as the griffin with a lion's body and an eagle's head and wings, and the Minotaur with a human body and a bull's head, were recorded in various murals and documents by people who lived in that era, and were conveyed along expanded trade routes for the sake of justifying the elite's status. The phenomenon of separating limbs or other features of various species and recombining them to create images of beings that correspond to the extended world that cannot be seen can be interpreted as a result of a change in society where the perspective of those who considered themselves "whole" centered on kinship relations expanded, leading to an increase in division of labor with people they had never met, and an awareness of themselves as "parts," which led to fear. In other words, one of the messages he wants to share is the discovery of the relationship between the creation of composite objects and self-awareness confirmed by technological advancements.

We have already seen countless interesting but grotesque images through AI image generation programs such as Midjourney and DALL-E since last year, such as a Roman-era statue of Spider-Man, an astronaut riding a horse on the moon's surface, and a robot painting with three arms, which go beyond the limits of the human body. The release of Sora will accelerate the spread of AI-generated videos even more. What is important is that the depiction and social transmission of monsters, i.e. composite objects, which are repeated along with the records of early human history, are being reproduced in the age of generative AI. This trend can be described as a "vibe" that envelops the world of this era, which is difficult to capture in a simple expression like trend.Although this expression is not new, we need to take this trend more seriously because we seem to be at a remarkable inflection point as we approach the AI era.

Long-form argumentation and explanation are impossible, and the most consumed and addictive YouTube and TikTok short videos prove the social consumption of decontextualized images.Momentary and associative poetic images have become increasingly important, which means that existing perspectives on interpretation related to what is considered information have been overturned.

In short, we need a change from what to see to how to see.

We need to choose a more semantic and analytical approach to trends and phenomena related to the body. It is an era where the philosophy of lingerie brands that confidently express a variety of bodies as they are is recognized instead of the aesthetics of the model legions in Guess and Calvin Klein marketing campaigns, and an image of pink lipstick above lips that does not look like a man or a woman becomes the embodiment of an effective strategy to approach teenagers. When imagining the future of technology and discussing technological advancements, the human body always exists alongside it. The question of how to view the human body will become even more complex and important in the future.

References

Creating video from text

Video generation models as world simulators

OpenAI’s DALL-E will train on Shutterstock’s library for six more years

The Origins of Monsters