Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety

👥 Fazl Barez, Tung-Yu Wu, Hunar Batra et al.

#MechanisticInterpretability #CausalCircuitDiscovery #SparseAutoencoders #TrainingDataAttribution

🔗 trendtoknow.com/papers/2611....

Training data attribution works without full model access using a proxy model—a surrogate trained on public data, or untrained models that provide influence signals. Read more: https://getnews.me/training-data-attribution-strategies-for-limited-model-access/ #trainingdataattribution #proxy