# PRISM: Probability Reallocation with In-Span Masking for Knowledge-Sensitive Alignment

arXiv: https://arxiv.org/abs/2604.01682 · HTML: https://mlsanity.com/paper/2604.01682/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 28 of its 34 references that are in the map.
- First posted on arXiv 2 Apr 2026.

## Builds on (most cited first, 28 of 28)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension](https://mlsanity.com/paper/1705.03551/) (2017)
- [Survey of Hallucination in Natural Language Generation](https://mlsanity.com/paper/2202.03629/) (2022)
- [Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection](https://mlsanity.com/paper/2310.11511/) (2023)
- [A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions](https://mlsanity.com/paper/2311.05232/) (2023)
- [FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation](https://mlsanity.com/paper/2305.14251/) (2023)
- [Instruction-Following Evaluation for Large Language Models](https://mlsanity.com/paper/2311.07911/) (2023)
- [Unsupervised Dense Information Retrieval with Contrastive Learning](https://mlsanity.com/paper/2112.09118/) (2021)
- [When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memories](https://mlsanity.com/paper/2212.10511/) (2022)
- [On Faithfulness and Factuality in Abstractive Summarization](https://mlsanity.com/paper/2005.00661/) (2020)
- [SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models](https://mlsanity.com/paper/2303.08896/) (2023)
- [LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset](https://mlsanity.com/paper/2309.11998/) (2023)
- [Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs](https://mlsanity.com/paper/2410.18451/) (2024)
- [Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?](https://mlsanity.com/paper/2405.05904/) (2024)
- [FLAME: Factuality-Aware Alignment for Large Language Models](https://mlsanity.com/paper/2405.01525/) (2024)
- [Understanding Finetuning for Factual Knowledge Extraction](https://mlsanity.com/paper/2406.14785/) (2024)
- [UAlign: Leveraging Uncertainty Estimations for Factuality Alignment on Large Language Models](https://mlsanity.com/paper/2412.11803/) (2024)
- [FactAlign: Long-form Factuality Alignment of Large Language Models](https://mlsanity.com/paper/2410.01691/) (2024)
- [Aligning Large Language Models to Follow Instructions and Hallucinate Less via Effective Data Filtering](https://mlsanity.com/paper/2502.07340/) (2025)
- [Smoothing Out Hallucinations: Mitigating LLM Hallucination with Smoothed Knowledge Distillation](https://mlsanity.com/paper/2502.11306/) (2025)
- [Beyond Under-Alignment: Atomic Preference Enhanced Factuality Tuning for Large Language Models](https://mlsanity.com/paper/2406.12416/) (2024)

## Abstract (arXiv)

Supervised fine-tuning (SFT) with token-level hard labels can amplify overconfident imitation of factually unsupported targets, causing hallucinations that propagate in multi-sentence generation. We study an augmented SFT setting in which training instances include coarse sentence-level factuality risk labels and inter-sentence dependency annotations, providing structured signals about where factual commitments are weakly supported. We propose PRISM, a differentiable risk-gated framework that modifies learning only at fact-critical positions. PRISM augments standard SFT with a lightweight, model-aware probability reallocation objective that penalizes high-confidence predictions on risky target tokens, with its scope controlled by span-level risk weights and model-aware gating. Experiments on hallucination-sensitive factual benchmarks and general evaluations show that PRISM improves factual aggregates across backbones while maintaining a competitive overall capability profile. Ablations further show that the auxiliary signal is most effective when used conservatively, and that knowledge masking and model-aware reallocation play complementary roles in balancing factual correction and capability preservation.