Technical News

OPENAI’S Sora 2 puts security and censorship to the test

Fees of a share of shares of $ 6.6 billion which made it the most precious private company in the world, the Tiktok style video application of Openai, fueled by its new model of artificial intelligence, Sora 2, becomes viral.

Despite the closed version which requires an invitation code, the video creation tool has already turned to the top of AppleApp Store and triggered a wave of Deepfake Fels, including a viral clip from CEO Sam Altman Gpus on display.

Internally, the deployment has rekindled a long -standing debate within Openai on how to balance security with freedom of creation.

A person familiar with the internal strategy of the company has declared that leadership considers strict as essential railing, but is also concerned with stifling creativity or being perceived as too much censorship.

This tension remains unresolved.

Openai culture has long favored speed, often shipping new tools before competitors and letting the public adapt in real time.

A former employee, who asked not to be appointed to discuss internal issues, told CNBC that during their mandate, Openai managers had a model for the prioritization of rapid launches. This strategy was fully exposed after China Deepseek published a powerful model at the end of last year which was cheaper and faster to build than anything that is Silicon Valley.

Openai responded in a few weeks, making her debut two new models in what was largely considered as a defensive decision to preserve his advance.

But Openai has a key advantage: its growing institutional muscle.

Formerly a disjointed research laboratory in the San Francisco mission district, the company has since been more structured, which allows it to run the interfunctional teams more quickly and to accelerate the cycles of development and deployment for products like Sora.

OPENAI said that Sora includes several layers of guarantees intended to prevent the dangerous content from being generated, using rapid filtering and output moderation between video frames and audio transcripts. It prohibits explicit content, terrorist propaganda and equipment promoting self -control. The application also uses filigranes and prohibitions for identifying the resemblance.

But some users have already found ways to get around these protections.

Sora 2, the AI ​​model fueling the OPENAI application, is a strong improvement compared to the first version. The new system generates longer and more coherent clips that seem surprisingly real.

Several viral videos present Altman after granting permission to his resemblance to be used on the platform, while others represent characters of popular cartoons like Pikachu and SpongeBob Squarepants in disturbing roles.

The content has fueled the criticisms that Openai moves again faster than its own railings. Its use of material protected by copyright – unless the rights holders oppose – comply with the current company’s current policy, although this approach is challenged in court.

Altman has rejected concerns, affirming in an article on X as Sora is as much a question of transparency – showing the public what technology can do – because it is a question of creating a commercial momentum to finance the wider ambitions of Openai around the general artificial intelligence.

The launch comes in the middle of the intensification of competition. Meta Launched the vibrations last week, a new video stream Ai short-circuit inside its Meta AI application. Google A Veo 3, while Bytedance and Alibaba also started rival systems.

OPENAI, on the other hand, is engaged in fresh expenses of $ 850 billion, deepening its thrust in infrastructure and new generation models.

Openai strikes the evaluation of $ 500 billion

Experts say that the push in the video is not only to attract more users in the ecosystem with another sticky consumption application.

Professor Hao Li, a leading expert in video synthesis, told CNBC that most AI systems are still trained on linguistic data such as books and internet text. But to evolve towards general intelligence, he said, the models must learn visual and audio information, a bit like a baby discovers the world through sight.

“We use AI to generate content to then form another model to work better,” he said.

Li added that its laboratory already uses a video generated by AI to improve model performance, strengthening synthetic data in the system.

This is part of a broader trend among researchers who see video generation as a way to simulate reality and help models to reason more like humans.

The former manager of Openai Zack Kass, whose next book “The next Renaissance: AI and the expansion of human potential” explore the societal implications of artificial intelligence, echoed from this point of view.

Regarding the question of how model manufacturers should tackle deployment, Kass argued that the compromises of liberation of powerful technologies are worth it.

“There are two alternatives to outdoor construction: not to build at all, or build in private. And these alternatives, for me, are worse,” he told CNBC. “If we have revolutionary technology, I think people should know and use it so that we can all update it.”

WATCH: OPENAI Ciments Status as the most precious private company in the world

OPENAI Ciments Status as the most precious private company in the world

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button