Incentivizing User Data Contributions for LLM Improvement under Withdrawal Rights

Di Feng, Chenhao Zhang, Zhanzhan Zhao

The continued improvement of large language models (LLMs) increasingly depends on eliciting high-quality, user-generated data, yet such data are costly to provide and often withhel…