Why did whisper take a million hours of YouTube videos?

According to the New York Times OpenAI, Google and Meta ignored corporate policies. They modified their own rules and discussed evading copyright law. The reason? The search for the information to train the newest artificial intelligence systems.

“In late 2021, OpenAI faced a supply problem” they report. There were no more reservoirs of reputable English-language text on the internet.  To train the next version of its technology they needed more data.

https://interestingengineering.com/culture/microsoft-and-openai-want-to-build-a-100-billion-datacenterhttps://interestingengineering.com/innovation/openais-new-voice-engine-clones-your-voice-in-only-15-secondshttps://interestingengineering.com/innovation/china-ubtech-chooses-chatgpt-rival-baidu-to-brain-power-humanoid-robot  

So what they did?

OpenAl researchers developed a speech recognition tool they named Whisper. The main purpose of the Whisper is to transcribe the audio from YouTube videos and make an A.I. smarter. As the New York Times reports a few OpenAI employees discussed how such a move might go …