AI developers are rapidly picking the Internet clean to train large language models such as those behind ChatGPT. Here’s how they are trying to get around the problem.

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