Approaching the Limits of AI Training Data

The expansion of artificial intelligence (AI) has been largely fueled by the vast amounts of data used to train neural networks, but experts warn that this growth is nearing its ceiling. A study indicated that by around 2028, the typical dataset size for training AI models will equal the total estimated amount of public online text. This suggests that the pool of conventional training data may be nearly depleted, which could hinder future advancements in AI technologies. Additionally, stricter regulations from content owners are limiting access to existing data, causing further concern regarding the data commons needed for ongoing AI development.