Model collapse – how LLMs become worse when trained on their own output i ibm.com 2 daymos 8 months ago 1 comments