Claim: OpenAI Used YouTube to Train GPT-4!

Although artificial intelligence models constantly leave our mouths open, there are some question marks that these tools bring with them. One of these is the data used for training. Use of data without permission infringement of copyrights It can cause

A report shared by The New York Times also draws attention to this point. According to the claim shared in the news, OpenAI is working to train the artificial intelligence model. He used Google data.

More than a million hours of YouTube videos were used to train GPT-4

NYT’s claim revealed that OpenAI benefited from a sizeable amount of YouTube data. Accordingly, the artificial intelligence giant, Whisper with a voice recognition tool called from a million hours transcribed many YouTube videos and compiled them using the most advanced language model. When training GPT-4 used.

In addition, the company knew that this situation could raise legal questions, but it won’t cause any problems It was also reported that he was thinking about it. It was claimed that Greg Brockman, who served as the president of the company, also took part in collecting the videos. The Times article adds that OpenAI exhausted the resources it used to train in 2021, and then began discussing its plan to transcribe YouTube content. Until then, the company had been using codes from Github, chess databases, and school content from Quizlet.

Matt Bryant, a spokesman for Google, which owns YouTube, told The Verge that he had seen “unconfirmed reports” on the issue and that such unauthorized uses Forbidden He stated that it was. Also, as we shared with you, a few days ago, YouTube CEO Neal Mohan announced that the platform’s using their data would be a violation he stated. Mohan, OpenAI’s new model Sora’s He made such a statement due to allegations that he was trained on YouTube.

Google itself trained models with YouTube data

Apart from these, there is information that Google itself collects data from YouTube. Spokesperson Bryant: In line with Google’s agreements with content producers to train their own models He stated that he used YouTube content. For this reason, it was also claimed that he did not take action against OpenAI.

All these claims reveal another face of artificial intelligence. Unauthorized use of data has the potential to create major problems with copyright infringement. We will wait and see what will happen regarding the issue.