model drift in LLMs

nlp
llm
why models degrade