This morning OpenAI was reported to be using transcripts of over 1 million hours of video from YouTube to train GPT-4 without the permission of Google or the content owners. Google also admits to doing the same but it is in line with YouTube's terms and conditions. But Google is also found to be using Google Docs and Google Sheets user data to train their AI.
The terms and conditions of Google's privacy policy were updated in July 2023 which allows user data of their services to be used for AI training purposes. Google however says this only happens if users give permission to be involved in testing new features.
To train the language model (LLM) that forms the basis of artificial intelligence (AI), trillions of quality data are required. What is happening now is a lack of data that will make LLM training more difficult in the near future even as the hardware for AI gets more powerful.
What may pass is that the AI is inadvertently trained using the work generated by generative AI. Like the Auroboros the head will eventually swallow the tail and future AI may not be able to reach the level of AGI as predicted.