Provable Training Data Identification for Large Language Models

Factual evidence

What the source reports

Researchers propose a provable set-level inference framework to statistically identify whether specific datasets were used to train an LLM.