Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs | Read Paper on Bytez