Check our new paper to revisit privacy, utility, and efficiency trade-offs when fine-tuning LLMs: ArXiv.