Research map / Large language models / Retrieval and knowledge
Copy-Paste to Mitigate Large Language Model Hallucinations
Yongchao Long, Yingying Zhang, Xianbin Wen, Xian Wu, Yuxi Zhou, Shenda Hong
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 |
|---|---|---|---|
| DPO on 365 high-copying preference samples | Llama-3-8B · FaithEval counterfactual | accuracy: 67.1 -> 92.8 | Improved |
| DPO on 365 high-copying preference samples | Llama-3.1-8B · FaithEval counterfactual | accuracy: 65.5 -> 92.6 | Improved |
| train on high-copying preferences (original context) | Mistral-7B-v0.2 · ConFiQA-MR original | accuracy: 71.20 -> 91.87 | Improved |
| train on high-copying preferences (original context) | Llama-3.1-8B · PubMedQA original | accuracy: 98.15 -> 97.67 | No clear effect |
| CopyPasteLLM (365 samples) vs Context-DPO (18,000 seen samples) | Llama-3-8B · ConFiQA-MR counterfactual | accuracy: 88.4 -> 80.9 | Hurt |
| CopyPasteLLM vs Context-DPO | Mistral-7B-v0.2 · ConFiQA-MC counterfactual | accuracy: 80.4 -> 82.5 | Improved |
| use CP-Order or CP-Refine prompting instead of Attributed | Llama-3.1-8B-Instruct · avg of RAGTruth, FaithEval, PubMedQA | faithfulness: 70.72 -> 91.39 (CP-Order) | Improved |
| CP-Link instead of CP-Order (ablation) | Llama-3.1-8B-Instruct · avg of three datasets | faithfulness: 91.39 -> 81.95 | Hurt |
| apply CopyPasteLLM on ambiguous (Negotiation) samples | Llama-3.1-8B · PubMedQA-Labeled Negotiation | accuracy: 46.38 -> 44.93 | No clear effect |
| apply CopyPasteLLM on clear-evidence (Consensus) samples | Llama-3-8B · PubMedQA-Labeled Consensus | accuracy: 75.49 -> 86.85 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ParamMute — ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- In-Context Sharpness as Alerts: An Inner Representation Perspective for Hallucination Mitigation (ICML 2024 · similarity 0.88)
- Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations (NeurIPS 2025 · similarity 0.87)
- The Labeling Problem in Hallucination Detection Benchmarks: An Empirical Evaluation (NeurIPS 2026 · similarity 0.87)
- ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability (ICLR 2025 · similarity 0.87)
- Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answer (NeurIPS 2026 · similarity 0.87)
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free Guarantees (ICML 2025 · similarity 0.87)
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.
