New technique traces AI behavior back to specific training data

Original: Studying Large Language Model Generalization With Influence Functions

Source: anthropic.com ↗