Shamraiz | December 15, 2024
OpenAI has found itself at the center of a heated debate over their video generator, Sora. The company showcased Sora in a video, leaving many tech enthusiasts and critics up in arms as the artificially generated AI produced a ‘chilling‘ gymnastic video that left it contemplating whether it actually knows any human physics or mechanics that govern its movement.
The Controversial Clip
The clip, which a text prompt produced, contained a gymnast who rotated through a sequence of apparently impossible stunts. It was far from elegant movement; it looked awkward, and even slightly frightening, at times her body acting as if she was actually defying physics and could fly away in the middle of a routine, it was like watching an exorcism.
This eerie outcome has fueled criticism of Sora’s capabilities. Many experts argue that the tool’s inability to mimic realistic human motion highlights significant limitations in AI’s progression toward understanding and replicating human-like intelligence.
Falling Short of the Turing Test
Critics were quick to point out that Sora’s performance does not pass the “Turing Test,” a benchmark used to evaluate whether a machine can exhibit human-like intelligence. The disturbing video underlines how much more generative AI needs to understand in complex human behaviors.
Gymnastics – The Ultimate AI Challenge
The video was shared on X by fintech professional Deedy Das, who has extensively experimented with text-to-video models. Das explained that gymnastics serves as a litmus test for generative AI due to its complex physics and intricate motion capture requirements.
“For the last six months, I’ve observed how text-to-video models struggle with complex physics movements like gymnastics,” Das told Ars Technica.
He shared additional examples of AI-generated gymnastics videos, noting that athletes sometimes appeared to split into two entities mid-performance. While acknowledging Sora’s improvements, such as fewer instances of teleporting gymnasts or mid-flip wardrobe changes, Das admitted the results remain “downright horrifying.”
“We hoped AI video would learn physics by default, but that hasn’t happened yet,” he added.
The Broader Implications of AI’s Struggles
Experts suggest that generative AI’s challenges extend beyond gymnastics. The technology also struggles with replicating natural language and comprehending basic human mechanics. One user aptly commented, “Watching Sora struggle with basic physics shows that simulating reality requires understanding it first.”
Another echoed this sentiment, stating, “We’re discovering that ‘common sense’ is anything but common – it’s one of the hardest things to encode.”
Progress Amidst Criticism
Despite the backlash, some acknowledge that Sora’s output represents progress. Just six months ago, AI-generated gymnastics clips were riddled with even more glaring errors. These advancements, though incremental, signal ongoing development in the field of generative AI. However, the road ahead remains long, as tools like Sora grapple with the intricacies of human motion and behavior.
