OpenAI has recently launched its new video application, Sora, introducing an opt-out policy for copyright holders. In its original framework, media companies were required to explicitly indicate their disapproval of AI-generated representations within the app. Following the release, the platform garnered attention for generating controversial content, including depictions of Nazi SpongeBob and criminal Pikachu, prompting OpenAI’s CEO, Sam Altman, to reassess the policy. Altman announced a shift, stating that the company would allow copyright holders to determine their preferred course of action.
During a discussion regarding the policy alteration, Altman indicated that the decision stemmed from dialogue with stakeholders, noting that he had not anticipated the level of backlash received.
“There was a difference between how we theorized it would feel for people and the actual responses once they were able to see the application in action,” Altman observed. He remarked that the feedback on images was particularly unexpected.
Sora offers a user experience similar to TikTok, enabling users to produce endless 10-second videos that include audio, featuring either AI-generated avatars or real individuals who have consented to be depicted. Although it aims to limit the inclusion of characters not registered in the app, users have been able to produce videos featuring copyrighted figures through the app’s text prompts.
Altman mentioned that numerous copyright holders expressed enthusiasm about Sora, yet called for “more controls” regarding their content. He acknowledged the application’s rapid popularity, explaining that the team had initially believed they could control its launch pace, which ultimately proved to be unfeasible.
“We genuinely care about the rights of content creators and users. Our goal is to implement additional controls, which will allow significant content pieces to be accessible with clear usage restrictions,” he added.
Altman remarked that the responses from early users showed a mix of feelings toward the idea of people utilizing their likenesses for AI-generated videos. While he expected users to either support or oppose the public use of their cameos, many expressed nuanced views. Consequently, OpenAI has begun to implement further restrictions, responding to users’ concerns about public representation and the potential for offensive content.
Bill Peebles, who oversees Sora, shared on social media that the team has received feedback from many users who wish to make their cameos available publicly while retaining control over their usage. Users are now able to provide specific instructions, such as avoiding political commentary or restricting the use of certain words.
Copyright holders are demanding ‘further controls’ on Sora
Peebles also noted that the team is improving the visibility of the Sora watermark on videos. Concerns have been raised about misinformation stemming from hyper-realistic AI-generated videos, particularly regarding the ease of removing less prominent watermarks.
“I am aware that some users are already discovering methods to erase the watermark,” Altman acknowledged during a recent question-and-answer session.
During a keynote at OpenAI’s DevDay, Altman announced the upcoming release of Sora 2 through OpenAI’s API, allowing developers to leverage the same underlying technology. This will enable them to create realistic AI-generated content without a watermark. Altman remained noncommittal when asked about implementing safeguards for the new API version.
Altman expressed surprise at the demand for using the app for generating videos intended for small, private groups, noting that although it has gained popularity, it does not align well with the app’s primary functionality.
He referred to the initial challenges as crucial learning experiences, emphasizing that competition in the space will soon increase. “Before long, we will not be the only effective video model available, and there will be videos generated without our protective measures, and that is part of the industry’s evolution,” Altman explained, stressing the need for societal readiness to engage with advanced video creation technology.
Altman indicated that there is a tendency for the public to overlook OpenAI’s technology during discussions, only paying attention when advancements are made. “We must foster a sort of co-evolution between technology and society,” he stated. “There will be clear challenges for society in adapting to enhanced video generation capabilities, but the best course of action to mitigate these issues is to let people experience the technology firsthand.”
“Challenges will arise as society adjusts to this technology”
This perspective may be contentious, particularly given AI’s long-standing impact on marginalized populations, often facilitating harm through various applications, from wrongful arrests to the dissemination of non-consensual explicit content. OpenAI has established certain restrictions for Sora; however, it faces challenges with individuals finding workarounds, as evidenced by the growing number of tutorials on watermark removal and unauthorized likeness generation within the app.
When questioned about whether OpenAI’s varied strategies imply a rapid, careless approach, Altman clarified that the current user feedback indicates that their restrictions are perceived by many as overly cautious and even censorial. He affirmed the company’s commitment to a measured rollout of features, stating that they will gradually expand capabilities in the future.
Turning a profit from Sora is not currently at the forefront of Altman’s concerns, although he acknowledged the necessity for profitability in the future. The focus now is on aggressive investments to ensure future success.
Despite the initial hurdles, OpenAI’s president, Greg Brockman, noted the swift adoption rate of Sora, surpassing even that of ChatGPT, maintaining a top position among free applications on Apple’s App Store. He indicated that this trend highlights the underlying need for computing resources moving forward, calling it a significant takeaway from the Sora launch.
Brockman hinted at the potential of Stargate, a collaboration between OpenAI, SoftBank, and Oracle, which aims to enhance AI infrastructure across the United States with an investment plan ranging from $100 billion to potentially $500 billion over four years. The venture has attracted attention from various stakeholders, including former President Donald Trump, amid discussions of new data centers in multiple states, despite controversies surrounding energy consumption and job creation promises.
OpenAI is pushing ahead with its plans, announcing a partnership with AMD that could lead to a 10 percent stake acquisition by OpenAI. Altman reaffirmed the company’s interest in exploring all aspects of AI infrastructure, indicating that more announcements would be forthcoming regarding their infrastructure development strategy.
During discussions, OpenAI executives repeatedly emphasized the pressing need for computational resources, highlighting how essential they are for scaling services effectively in the competitive landscape.
Brockman likened the demand for compute resources to inquiries about workforce size, suggesting that increased capacity will be necessary to maximize output. As the application landscape evolves, the capacity to create hyper-realistic videos remains a focal point for users and developers alike.
