Search papers, labs, and topics across Lattice.
1
2
0
Training methods reshape refusal mechanisms in language models, but no single approach achieves the trifecta of robustness, capability, and correctability.
