Search papers, labs, and topics across Lattice.
Chung-Ang University
1
0
2
LLMs can be steered to respect instruction hierarchies in multi-turn dialogues without the need for costly fine-tuning, revealing a novel approach to mitigating role-influence inversion.
