1 paper · 1 filter
Isaac Picov, Ritesh Goru
Explaining closed-source Large Language Model (LLM) outputs is challenging because API access prevents gradient-based attribution, while perturbation methods are costly and noisy w…