Google Allowed OpenAI to Use 1 Million Hours of YouTube Video to Train its GPT-4 Model
It is claimed that more than 1 million hours of YouTube videos were copied by OpenAI in order to train its most advanced large language model (LLM), which it calls GPT-4.
According to a New York Times report, OpenAI developed the Whisper audio transcription model that helps the company collect data from YouTube videos. The publication stated that OpenAI knew this method might come under scrutiny, but they opted for this method because they believed it was fair use. Interestingly, Google, which owns YouTube, is also alleged to be doing the same for its AI models, thus violating its creator’s copyright.
New York Times news, highlighting the claim that OpenAI extracted data from YouTube videos and podcasts to train two artificial intelligence systems The InformationIn line with the news of.
YouTube CEO Neil Mohan Bloomberg When interviewed by , he said the company’s policies “do not allow downloading things like transcripts or video snippets, and that’s a clear violation of our terms of service.” On the other hand, when asked whether YouTube data was used by OpenAI, Mohan gave a vague answer and said, “I have seen reports that it may or may not have been used. “I don’t have any information either,” he said.
It is claimed that some people at Google knew about OpenAI’s practice of copying YouTube data, but they could not do anything because Google also resorted to the same practice to train its own artificial intelligence model. Google, on the other hand, said that it only scrapes data from videos after the creator of the video gives permission.
According to the report, it is claimed that Google asked a team to “change its privacy policy” in June 2023, “so that Google can leverage public Google Docs, restaurant reviews on Google Maps, and other online materials for more AI products.”
It seems that many more questions will arise about the methods used to train artificial intelligence models. What do you think?
