# Teaching with Lies: Curriculum DPO on Synthetic Negatives for Hallucination Detection

arXiv: https://arxiv.org/abs/2505.17558 · HTML: https://mlsanity.com/paper/2505.17558/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion and multimodal model alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 18 of its 30 references that are in the map.
- First posted on arXiv 23 May 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)

## Builds on (most cited first, 18 of 18)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Large Language Models Encode Clinical Knowledge](https://mlsanity.com/paper/2212.13138/) (2022)
- [PubMedQA: A Dataset for Biomedical Research Question Answering](https://mlsanity.com/paper/1909.06146/) (2019)
- [BLEURT: Learning Robust Metrics for Text Generation](https://mlsanity.com/paper/2004.04696/) (2020)
- [DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs](https://mlsanity.com/paper/1903.00161/) (2019)
- [FEQA: A Question Answering Evaluation Framework for Faithfulness Assessment in Abstractive Summarization](https://mlsanity.com/paper/2005.03754/) (2020)
- [MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documents](https://mlsanity.com/paper/2404.10774/) (2024)
- [A Token-level Reference-free Hallucination Detection Benchmark for Free-form Text Generation](https://mlsanity.com/paper/2104.08704/) (2021)
- [Large Language Models in Law: A Survey](https://mlsanity.com/paper/2312.03718/) (2023)
- [Curriculum Learning for Domain Adaptation in Neural Machine Translation](https://mlsanity.com/paper/1905.05816/) (2019)
- [Simple and Effective Curriculum Pointer-Generator Networks for Reading Comprehension over Long Narratives](https://mlsanity.com/paper/1905.10847/) (2019)
- [Lynx: An Open Source Hallucination Evaluation Model](https://mlsanity.com/paper/2407.08488/) (2024)
- [Curry-DPO: Enhancing Alignment using Curriculum Learning & Ranked Preferences](https://mlsanity.com/paper/2403.07230/) (2024)
- [Constrained Decoding for Neural NLG from Compositional Representations in Task-Oriented Dialogue](https://mlsanity.com/paper/1906.07220/) (2019)
- [Using LLMs in Software Requirements Specifications: An Empirical Evaluation](https://mlsanity.com/paper/2404.17842/) (2024)
- [FactCheckmate: Preemptively Detecting and Mitigating Hallucinations in LMs](https://mlsanity.com/paper/2410.02899/) (2024)

## Built on by (most cited first, 3 of 3)

- [HalluGuard: Evidence-Grounded Small Reasoning Models to Mitigate Hallucinations in Retrieval-Augmented Generation](https://mlsanity.com/paper/2510.00880/) (2025)
- [V-Loop: Visual Logical Loop Verification for Hallucination Detection in Medical Visual Question Answering](https://mlsanity.com/paper/2601.18240/) (2026)
- [Uncertainty-Aware Web-Conditioned Scientific Fact-Checking](https://mlsanity.com/paper/2604.11036/) (2026)

## Abstract (arXiv)

Aligning large language models (LLMs) to accurately detect hallucinations remains a significant challenge due to the sophisticated nature of hallucinated text. Recognizing that hallucinated samples typically exhibit higher deceptive quality than traditional negative samples, we use these carefully engineered hallucinations as negative examples in the DPO alignment procedure. Our method incorporates a curriculum learning strategy, gradually transitioning the training from easier samples, identified based on the greatest reduction in probability scores from independent fact checking models, to progressively harder ones. This structured difficulty scaling ensures stable and incremental learning. Experimental evaluation demonstrates that our HaluCheck models, trained with curriculum DPO approach and high quality negative samples, significantly improves model performance across various metrics, achieving improvements of upto 24% on difficult benchmarks like MedHallu and HaluEval. Additionally, HaluCheck models demonstrate robustness in zero-shot settings, significantly outperforming larger state-of-the-art models across various benchmarks.