AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Breaking the Trade-Off Between Faithfulness and Expressiveness for Large Language Models

Chenxu Yang, Qingyi Si, Lanrui Wang, Zheng Lin

AAAI 2026 · Large language models › Retrieval and knowledge · approach: Knowledge editing in LLMs · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add knowledge-aware reranking (ablation)CoDe · FAITHDIALH-Judge: 85.5 -> 90.8Improved
add knowledge-aware reranking (ablation)CoDe · FAITHDIALCRE: 33.9 -> 29.9Hurt
add adaptive fusion weight alpha (ablation)CoDe · FAITHDIALAvg. of all metrics: 51.4 -> 52.2Improved
add expressiveness-oriented stream (ablation)CoDe · FAITHDIALCRE: 27.3 -> 29.9Improved
add semantic reward in reranking (ablation)CoDe · FAITHDIALH-Judge: 88.6 -> 90.8Improved
add attentive reward in reranking (ablation)CoDe · FAITHDIALH-Judge: 88.3 -> 90.8Improved
use CoDe instead of greedy decodingLlama2-7B-chat · FAITHDIALH-Judge: 86.9 -> 90.8Improved
use CoDe instead of greedy decodingQwen-2.5-3B-chat · FAITHDIALH-Judge: 90.5 -> 92.9Improved
use stochastic sampling (top-k) instead of greedyLlama2-7B-chat · FAITHDIALH-Judge: 86.9 -> 75.3Hurt
use CAD instead of greedyLlama2-7B-chat · FAITHDIALCRE: 30.0 -> 21.7Hurt
use CoDe instead of greedyQwen-2.5-14B-chat · FAITHDIALCOH: 53.6 -> 53.6No clear effect
use CoDe instead of greedyLlama2-7B-chat · NQ, NQ-SWAP, HalluEval (QA)Avg. accuracy: 37.9 -> 41.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy