Search papers, labs, and topics across Lattice.
1
0
2
Integrating Sparse Autoencoders into transformer models can slash jailbreak success rates by up to 5x, reshaping our understanding of model robustness against adversarial attacks.
