Researchers warn that companies are rapidly depleting high-quality human-written training data for AI models. This data bottleneck could halt advancements in AI capabilities.

• Feast or Famine: AI thrives on vast amounts of data, often harvested from public online sources. Legal battles with publishers highlight the ethical concerns around this practice. As companies automate tasks, the flow of new training data might decrease.

• Running Out of Fresh Food: A new study predicts that the growth rate of text data for training AI falls short of computing power advancements. This suggests large language models could exhaust usable data by 2026.

• AI Ouroboros: Self-Eating Solution? In the absence of fresh data, companies might turn to AI-generated content for training. However, experts warn this "synthetic data" could lead to a decline in output quality, creating a loop where AI consumes its own inferior creations.

"I think it's important to keep in mind that we don't necessarily need to train larger and larger models," AI researcher and University of Toronto assistant professor of computer engineering Nicolas Papernot, who was not involved in the study, told the AP.