Patrick Leask and Noura Al Moubayed introduce SAE stitching, a new method for mechanistic intepretability, in a poster at NeurIPS 2024.
AI Forensics © 2025
