Search Results for author:
Found 2 papers, 0 papers with code
Position Paper: Toward New Frameworks for Studying Model Representations
Mechanistic interpretability (MI) aims to understand AI models by reverse-engineering the exact algorithms neural networks learn.
An Adversarial Example for Direct Logit Attribution: Memory Management in gelu-4l
That is, clearing residual stream directions set by earlier layers by reading in information and writing out the negative version.
Cannot find the paper you are looking for? You can
Submit a new open access paper.
