1Model collapse – how LLMs become worse when trained on their own output (opens in new tab)(ibm.com)2daymos1mo ago1