Beware of AI ‘model collapse’: How training on synthetic data pollutes the next generation

Oxford scholars found that large language models fed a diet of 'cannibal' data, created by other LLMs, sink into complete gibberish.

Source: Latest news