New method identifies which training data shaped AI model outputs

Original: Data Attribution in Large Language Models via Bidirectional Gradient Optimization

Source: arxiv.org ↗