Abstract
A central challenge in modern language models (LMs) is intrinsic hallucination: the generation of information that is plausible but unsubstantiated relative to input context. To study this problem, we propose Precise Information Control (PIC), a new task formulation that requires models to generate long-form outputs grounded in a provided set of short self-contained statements, known as verifiable claims, without adding any unsupported ones. For comprehensiveness, PIC includes a full setting that tests a model's ability to include exactly all input claims, and a partial setting that requires the model to selectively incorporate only relevant claims. We present PIC-Bench, a benchmark of eight long-form generation tasks (e.g., summarization, biography generation) adapted to the PIC setting, where LMs are supplied with well-formed, verifiable input claims. Our evaluation of a range of open and proprietary LMs on PIC-Bench reveals that, surprisingly, state-of-the-art LMs still intrinsically hallucinate in over 70% of outputs. To alleviate this lack of faithfulness, we introduce a post-training framework, using a weakly supervised preference data construction method, to train an 8B PIC-LM with stronger PIC ability--improving from 69.1% to 91.0% F1 in the full PIC setting. When integrated into end-to-end factual generation pipelines, PIC-LM improves exact match recall by 17.1% on ambiguous QA with retrieval, and factual precision by 30.5% on a birthplace verification task, underscoring the potential of precisely grounded generation.
Abstract (translated)
现代语言模型（LM）的一个核心挑战是内在幻觉：生成的信息在输入上下文中看似合理但缺乏实质证据支持。为研究这一问题，我们提出了精确信息控制（PIC），这是一种新的任务形式化方法，要求模型基于一组提供的简短独立陈述——称为可验证声明——生成长文本输出，并且不允许添加任何未被证实的信息。为了全面评估，PIC 包含一个完整的设置来测试模型能否准确地包括所有的输入声明，以及一个部分设置以要求模型选择性地整合相关的声明。我们推出了 PIC-Bench，这是一个由八个改编至 PIC 设置的长文本生成任务（例如总结、人物生成）构成的基准，其中语言模型接收结构良好且可验证的输入陈述。 我们在 PIC-Bench 上评估了一系列开源和专有语言模型后发现，令人惊讶的是，最先进的语言模型仍然在超过 70% 的输出中产生内在幻觉。为了解决这一忠实度不足的问题，我们引入了一个后训练框架，利用弱监督偏好数据构造方法来训练一个具有更强 PIC 能力的8B参数PIC-LM——在完整的PIC设置下F1得分从69.1%提升到91.0%。 将PIC-LM集成进端到端事实生成流水线中，在带有检索的模糊问题回答任务中，精确匹配召回率提升了17.1%，而在出生地验证任务中，则实现了30.5%的事实精度提升。这些结果强调了精准基础性生成的潜力。
URL
https://arxiv.org/abs/2506.06589
