Search papers, labs, and topics across Lattice.
Xidian University
1
0
3
LLMs can be steered to generate more faithful reasoning chains without sacrificing accuracy using a novel geometric and entropy-based framework, outperforming even GPT-4 in faithfulness detection.
