Can We Infer Confidential Properties of Training Data from LLMs? | Read Paper on Bytez