Why Sora Uses Videos from Vimeo, YouTube, and 18+ Sites
Posted: Sun Jan 19, 2025 4:08 am
OpenAI CTO Mira Murati confirmed that the neural network will be trained on videos that are currently in the public domain. However, she declined to name specific video services or sites.
The main part of the training sample of any neural costa rica whatsapp number data network is the content of web pages and books. The content of user dialogues with neural networks is not always taken into account in training and only to maintain the dialogue.
As is known, owners of popular websites are already forced to regularly contact OpenAI to prohibit the use of their own content for training models. The New York Times even went to court on this matter. In Russia, Roskomnadzor sent special recommendations to owners of communication networks on how to close websites from OpenAI.
Obviously, Sora will be trained on videos posted on the two most popular video services - Vimeo and YouTube. Focusing on small sites does not make sense due to the peculiarities of the video player used in each individual case.
The terms of OpenAI's agreements with video services will likely not be disclosed. And even the user agreements of video services with creators who monetize their content through popular platforms will contain vague wording.
Porn sites are also of interest for training AI models. And the developer of the neural network will also need to negotiate with the owners of such sites. Adult content is used so that the neural network learns what naked human figures look like in realistic proportions.
The main part of the training sample of any neural costa rica whatsapp number data network is the content of web pages and books. The content of user dialogues with neural networks is not always taken into account in training and only to maintain the dialogue.
As is known, owners of popular websites are already forced to regularly contact OpenAI to prohibit the use of their own content for training models. The New York Times even went to court on this matter. In Russia, Roskomnadzor sent special recommendations to owners of communication networks on how to close websites from OpenAI.
Obviously, Sora will be trained on videos posted on the two most popular video services - Vimeo and YouTube. Focusing on small sites does not make sense due to the peculiarities of the video player used in each individual case.
The terms of OpenAI's agreements with video services will likely not be disclosed. And even the user agreements of video services with creators who monetize their content through popular platforms will contain vague wording.
Porn sites are also of interest for training AI models. And the developer of the neural network will also need to negotiate with the owners of such sites. Adult content is used so that the neural network learns what naked human figures look like in realistic proportions.