Search papers, labs, and topics across Lattice.
2
0
5
SAFARI maintains diagnostic precision even when critical fault information lies five times beyond the model's context limits, a feat traditional methods cannot achieve.
LLMs can be made significantly more robust to jailbreaks by weighting the reasoning steps in DPO training, leading to more principled refusals.
