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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add knowledge-aware reranking (ablation) | CoDe · FAITHDIAL | H-Judge: 85.5 -> 90.8 | Improved |
| add knowledge-aware reranking (ablation) | CoDe · FAITHDIAL | CRE: 33.9 -> 29.9 | Hurt |
| add adaptive fusion weight alpha (ablation) | CoDe · FAITHDIAL | Avg. of all metrics: 51.4 -> 52.2 | Improved |
| add expressiveness-oriented stream (ablation) | CoDe · FAITHDIAL | CRE: 27.3 -> 29.9 | Improved |
| add semantic reward in reranking (ablation) | CoDe · FAITHDIAL | H-Judge: 88.6 -> 90.8 | Improved |
| add attentive reward in reranking (ablation) | CoDe · FAITHDIAL | H-Judge: 88.3 -> 90.8 | Improved |
| use CoDe instead of greedy decoding | Llama2-7B-chat · FAITHDIAL | H-Judge: 86.9 -> 90.8 | Improved |
| use CoDe instead of greedy decoding | Qwen-2.5-3B-chat · FAITHDIAL | H-Judge: 90.5 -> 92.9 | Improved |
| use stochastic sampling (top-k) instead of greedy | Llama2-7B-chat · FAITHDIAL | H-Judge: 86.9 -> 75.3 | Hurt |
| use CAD instead of greedy | Llama2-7B-chat · FAITHDIAL | CRE: 30.0 -> 21.7 | Hurt |
| use CoDe instead of greedy | Qwen-2.5-14B-chat · FAITHDIAL | COH: 53.6 -> 53.6 | No clear effect |
| use CoDe instead of greedy | Llama2-7B-chat · NQ, NQ-SWAP, HalluEval (QA) | Avg. accuracy: 37.9 -> 41.1 | Improved |
Closest papers by meaning
- Speculate, then Collaborate: Fusing Knowledge of Language Models during Decoding (ICML 2025 · similarity 0.89)
- ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation (NeurIPS 2025 · similarity 0.88)
- To Trust or Not to Trust? Enhancing Large Language Models' Situated Faithfulness to External Contexts (ICLR 2025 · similarity 0.88)
- A Probabilistic Perspective on Unlearning and Alignment for Large Language Models (ICLR 2025 · similarity 0.88)
- KVSmooth: Mitigating Hallucination in Multi-modal Large Language Models through Key-Value Smoothing (CVPR 2026 · similarity 0.88)
- DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models (ICLR 2024 · similarity 0.88)
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.
