论文检索

Applications · Language, Speech and Dialog

Chung-Ming Chien, Manu Orsini, Eugene Kharitonov, Neil Zeghidour, Karen Livescu, Alexandre Défossez

Speech-to-speech language models have recently emerged to enhance the naturalness of conversational AI. In particular, full-duplex models are distinguished by their real-time interactivity, including handling of pauses, interruptions, and backchannels. However, improving their factuality remains an open challenge. While scaling the model size could address this gap, it would make real-time inference prohibitively expensive. In this work, we propose Moshi-RAG, a modular approach that combines a compact full-duplex interface with selective retrieval to access more powerful knowledge sources. Our asynchronous framework enables the model to identify knowledge-demanding queries and ground its responses in external information. By leveraging the natural temporal gap between response onset and the delivery of core information, the retrieval process can be completed while maintaining a natural conversation flow. With this approach, Moshi-RAG achieves factuality comparable to the best publicly released non-duplex speech language models while preserving the interactivity inherent to full-duplex systems. Moreover, our flexible design supports plug-and-play retrieval methods without retraining and demonstrates strong performance on out-of-domain mathematical reasoning tasks.

Deep Learning · Large Language Models

Shankar Padmanabhan, Mustafa Omer Gul, Tanya Goyal

Post-training endows pretrained LLMs with a variety of desirable skills, such as instruction-following, reasoning, and others. However, these post-trained LLMs only encode knowledge up to a cut-off date, necessitating continual adaptation. Unfortunately, existing solutions cannot effectively learn new knowledge from adaptation document corpora and simultaneously mitigate the forgetting of earlier learned capabilities. To address this, we introduce Distillation via Split Contexts (DiSC), a simple context-distillation based approach for continual knowledge adaptation. DiSC derives student and teacher distributions by conditioning on distinct segments of the training example and minimizes the KL divergence between them for the common tokens. This insight allows us to efficiently apply context-distillation without requiring explicit generation steps during training. We run experiments on three post-trained models and two adaptation domains. Compared to prior finetuning and distillation methods for continual adaptation, DiSC consistently reports the best trade-off between learning new knowledge and mitigating forgetting of previously learned skills like instruction-following and reasoning, or factual knowledge.

Applications · Computer Vision

Zhaoyang Li, Zhichao You, Tianrui Li

Although multi-modal learning has advanced point cloud completion, the theoretical mechanisms remain unclear. Recent works attribute success to the connection between modalities, yet we identify that standard hard projection severs this connection, inducing Cross-Modal Entropy Collapse where sparse support hinders visual prior propagation. To bridge this gap, we propose SplAttN, which maximizes Point-wise Mutual Information via Differentiable Gaussian Splatting. By reformulating projection as continuous density estimation, SplAttN facilitates gradient flow and optimizes connection learnability. Extensive experiments show that SplAttN achieves state-of-the-art performance on PCN and ShapeNet-55/34. Crucially, we utilize the real-world KITTI benchmark as a stress test for multi-modal reliance. Counter-factual evaluation reveals that while baselines degenerate into unimodal template retrievers insensitive to visual removal, SplAttN maintains a robust dependency on visual cues, validating that our method establishes an effective cross-modal connection. Code is available at https://anonymous.4open.science/r/Anonymous-766B/.

Deep Learning · Large Language Models

Itamar Pres, Belinda Li, Laura Ruis, Carl Guo, Keya Hu, Mehul Damani, Isha Puri, Ekdeep Singh Lubana, Jacob Andreas

Despite ever-increasing sophistication in language model (LM) pre- and post-training pipelines, many important failures persist: models overcondition on user framing (“sycophancy”), exhibit incomplete logical generalization, and produce confident but incorrect responses. We argue that these failures arise from a modeling assumption permeating all aspects of the pipeline: that behavior can be specified and evaluated independently on single-output pairs. Many model failures are difficult, if not impossible, to detect without reasoning about relationships between a model’s responses across inputs. In this position paper, we propose self-consistency as a framework for understanding these failures. We first observe that a wide variety of techniques designed to improve specific aspects of LM behavior—targeting properties as diverse as adversarial robustness and factual coherence—can be understood as special cases of a common “consistency optimization” procedure and addressed with a standard set of optimization tools. We next outline a set of new model properties that could be achieved by optimizing for consistency, and conclude with a discussion of what it would mean to develop generally consistent LMs, including the capabilities they would enable and the objections they raise.

Deep Learning · Large Language Models

Zhepei Wei, Xiao Yang, Kai Sun, Jiaqi Wang, Rulin Shao, Jingxiang Chen, Mohammad Kachuee, Teja Gollapudi, Yiwei Liao, Nicolas SCHEFFER 等

While large language models (LLMs) have demonstrated strong performance on factoid question answering, they are still prone to hallucination and untruthful responses, particularly when tasks demand information outside their parametric knowledge. Indeed, truthfulness requires more than accuracy---models must also recognize uncertainty and abstain when unsure to avoid hallucinations. This presents a fundamental challenge for existing methods: approaches that optimize for accuracy often amplify hallucinations, while those that encourage abstention can become overly conservative, sacrificing correct answers. Both extremes ultimately compromise truthfulness. In this work, we present TruthRL, a general reinforcement learning (RL) framework that directly optimizes the truthfulness of LLMs. Specifically, we implement TruthRL using GRPO with a simple yet effective ternary reward that distinguishes correct answers, hallucinations, and abstentions. It incentivizes models to reduce hallucinations not only by providing correct responses, but also by enabling abstention when uncertain, thereby improving truthfulness. Extensive experiments across four knowledge-intensive benchmarks show that \model significantly reduces hallucinations (e.g., 43.5\% $\rightarrow$ 19.4\%) and improves truthfulness (e.g., 5.3\% $\rightarrow$ 37.2\%), with consistent gains across various backbone models (e.g., Qwen, Llama). In-depth ablation study demonstrates that vanilla accuracy-driven methods such as supervised fine-tuning or RL with a binary reward struggle to balance factual correctness and uncertainty, whereas the truthfulness-driven TruthRL achieves strong performance in both accuracy and truthfulness, underscoring the importance of learning objective design for developing truthful LLMs. Moreover, we find the improvement of \model arises from enhancing the capability of LLMs to recognize their knowledge boundary, hence avoiding being overly conservative as the baselines are. Further analysis validates our method across multiple evaluation judges, and confirms that TruthRL is robust to hallucination-baiting questions.

Deep Learning · Large Language Models

Ziqiang Cui, Yunpeng Weng, Xing Tang, Peiyang Liu, Shiwei Li, Bowei He, Jiamin Chen, Yansen Zhang, xiuqiang He, Rui Zhang 等

Retrieval-Augmented Generation (RAG) has emerged as a promising paradigm for improving the timeliness of knowledge updates and the factual accuracy of large language models. However, incorporating a large volume of retrieved documents significantly increases input length, leading to prohibitive computational costs. Existing compression approaches often compromise task performance, primarily due to their reliance on predefined heuristics. These heuristics fail to ensure that the compressed context is conducive to the generation tasks. To address these limitations, we propose CORE-RAG, a novel framework for context compression in RAG systems. CORE eliminates reliance on proxy heuristics through a performance-driven learning framework, which directy utilizes task performance as a feedback signal to iteratively refine the compressor policy. Prior to this optimization process, we incorporate a knowledge distillation phase to initialize the compressor with a robust policy. Extensive experiments demonstrate the superiority of our approach. At a high compression ratio of 3\%, CORE not only avoids performance degradation but also improves the average Exact Match (EM) score by 3.3 points compared to using full documents. Our code is available at https://anonymous.4open.science/r/CORE-28B4.

Deep Learning · Large Language Models

Yiqun Yao, Xiang Li, Xin Jiang, Xuezhi Fang, Naitong Yu, Siwei Dong, Wenjia Ma, Jing Li, Aixin Sun, Yequan Wang

It is widely recognized that, after generative pre-training, Transformer FeedForward layers implicitly function as semantic memory, encoding linguistic and factual knowledge, while the contexts in key–value (KV) cache contain raw events, serving as the source of models' episodic memory. In this work, we show that a same group of Transformer FeedForward-layer parameters can both be semantic and episodic memory, which is retrievable without explicitly attending to the related KV cache. To realize this idea, we introduce Hypermem, a hypernetwork that recurrently maps contexts into targeted updates of FeedForward parameters. We post-train the hypernetwork using continuation and random-access associative memory objectives, eliminating the need for test-time training. Extensive experiments demonstrate that our approach outperforms related methods, including MemoryLLM and generative adapter, on memory retrieval, long-context question answering, and personalization benchmarks, establishing a new state of the art for hypernetwork-based memory mechanisms. Our results suggest that directly bridging data and parameters provides a viable direction for exploring next-generation foundation models with more flexible and persistent memory capabilities.

Applications · Health / Medicine

Akash Ghosh, Srivarshinee Sridhar, Raghav Ravi, Muhsin Muhsin, Sriparna Saha, Chirag Agarwal

Integrating language models (LMs) in healthcare systems holds great promise for improving medical workflows and decision-making. However, a critical barrier to their real-world adoption is the lack of reliable evaluation of their trustworthiness, especially in multilingual healthcare settings. Existing LMs are predominantly trained in high-resource languages, making them ill-equipped to handle the complexity and diversity of healthcare queries in mid- and low-resource languages, posing significant challenges for deploying them in global healthcare contexts where linguistic diversity is key. In this work, we present \textsc{Clinic}, a \textbf{C}omprehensive Mu\textbf{l}tilingual Benchmark to evaluate the trustworth\textbf{i}ness of la\textbf{n}guage models \textbf{i}n health\textbf{c}are. \name systematically benchmarks LMs across five key dimensions of trustworthiness: truthfulness, fairness, safety, robustness, and privacy, operationalized through 18 diverse tasks, spanning 15 languages (covering all the major continents), and encompassing a wide array of critical healthcare topics like disease conditions, preventive actions, diagnostic tests, treatments, surgeries, and medications. Our extensive evaluation reveals that LMs struggle with factual correctness, demonstrate bias across demographic and linguistic groups, and are susceptible to privacy breaches and adversarial attacks. By highlighting these shortcomings, \name lays the foundation for enhancing the global reach and safety of LMs in healthcare across diverse languages.

Deep Learning · Large Language Models

Xilun Chen, Ilia Kulikov, Vincent-Pierre Berges, Barlas Oğuz, Rulin Shao, Gargi Ghosh, Scott Yih

Reasoning Large Language Models (R-LLMs) have significantly advanced complex reasoning tasks but often struggle with factuality, generating substantially more hallucinations than their non-reasoning counterparts on long-form factuality benchmarks. However, extending online Reinforcement Learning (RL), a key component in recent R-LLM advancements, to the long-form factuality setting poses several unique challenges due to the lack of reliable verification methods. Previous work has utilized automatic factuality evaluation frameworks such as FActScore to curate preference data in the offline RL setting, yet we find that directly leveraging such methods as the reward in online RL leads to reward hacking in multiple ways, such as producing less detailed or relevant responses. We propose a novel reward function that simultaneously considers the factual precision, response detail level, and answer relevance, and applies online RL to learn high quality factual reasoning. Evaluated on six long-form factuality benchmarks, our factual reasoning model achieves an average reduction of 23.1 percentage points in hallucination rate, a 23% increase in answer detail level, and no degradation in the overall response helpfulness.

General Machine Learning · Evaluation

Xin Gao, Cheng Yang, Chufan Shi, Taylor Berg-Kirkpatrick

Unified multimodal models (UMMs) emerge as a promising paradigm for general-purpose multimodal intelligence. As they are deployed in real-world applications, effectively updating internal knowledge becomes critical. While knowledge editing methods have matured for text-only models, a fundamental question remains unexplored: do knowledge edits that successfully modify textual outputs transfer to image generation for UMMs? To this end, we introduce UniKE, the first benchmark for cross-modality knowledge editing in UMMs, comprising 3,005 instances across attribute edits and relation edits. We propose an automated VQA-based evaluation protocol to assess factual consistency between edited knowledge and generated images. Our evaluation reveals a striking modality gap: parameter-editing methods achieving high text-side efficacy (up to 93\%) fail to produce visual changes, with VQA accuracy below 6\% under direct generation. We propose Reasoning-augmented Parameter Editing, which explicitly activates edited knowledge before generation, improving visual verification to 10-27\% for attributes. Through mechanistic analysis, we identify the root cause: edit-affected pathways exhibit near-random overlap with visual attribute-conditioning channels, indicating a fundamental pathway mismatch. These findings demonstrate that textual knowledge edits do not guarantee cross-modality transfer, motivating future work on modality-aware editing methods.

Social Aspects · Privacy

Seongho Joo, Hyukhun Koh, Kyomin Jung

Large language models (LLMs) often employ Retrieval-Augmented Generation (RAG) to improve factuality. However, this also increases the risk of sensitive private information leakage. Differential Privacy (DP) has therefore been integrated into LLM inference and is widely regarded as a standard safeguard; yet most studies focus narrowly on the privacy–utility trade-off, leaving the trustworthiness of DP outputs underexplored. To assess trustworthiness, we employ the confidence gap (CG), which quantifies an LLM’s internal knowledge conflict. We show that CG correlates with both hallucination and exposure of personally identifiable information (PII). Building on this insight, we propose PEARL, a CG‑guided, entropy‑aware private decoding framework. PEARL adaptively allocates the privacy budget across tokens and sentences based on CG, concentrating protection on PII-bearing spans while stabilizing low-confidence, hallucination-prone regions. In experiments, PEARL improves both trustworthiness and robustness against PII extraction attacks. Notably, while applying DP alone significantly increases hallucination, our framework demonstrates that it is possible to preserve privacy while reducing hallucination.

Social Aspects · Safety

Lin Li, Georgia Channing, Suhaas Bhat, Gabriel Jones, Yarin Gal

Large language models (LLMs) have achieved remarkable progress in open-ended text generation, yet they remain prone to hallucinating incorrect or unsupported content, which undermines their reliability. This issue is exacerbated in long-form generation due to hallucination snowballing, a phenomenon where early errors propagate and compound into subsequent outputs. To address this challenge, we propose a novel inference-time hallucination mitigation framework, named Segment-wise HAllucination Rejection Sampling (SHARS), which uses am arbitrary hallucination detector to identify and reject hallucinated segments during generation and resample until faithful content is produced. By retaining only confident information and building subsequent generations upon it, the framework mitigates hallucination accumulation and enhances factual consistency. To instantiate this framework, we adopt semantic uncertainty as the detector and introduce several vital modifications to address its limitations and better adapt it to long-form text. Our method enables models to self-correct hallucinations without requiring external resources such as web search or knowledge bases, while remaining compatible with them for future extensions. Empirical evaluations on standardized hallucination benchmarks demonstrate that our method substantially reduces hallucinations in long-form generation while preserving or even improving the informativeness of generation.

Applications · Social Sciences

Jiahao Qiu, Fulian Xiao, Yimin Wang, Yuchen Mao, Yijia Chen, Xinzhe Juan, Siran Wang, Xuan Qi, Tongcheng Zhang, Zixin Yao 等

Recent advances in large language models (LLMs) have led to remarkable progress across various domains, yet their capabilities in the humanities, particularly history, remain underexplored. Historical reasoning poses unique challenges for LLMs, involving multimodal source interpretation, temporal inference, and cross-linguistic analysis. Existing general-purpose agents perform well on many current benchmarks but lack the domain expertise needed to address complex historical questions. To address this gap, we introduce HistBench, a new benchmark of 414 high-quality and carefully-reviewed questions stratified by difficulty and designed to evaluate LLM's capacity for historical reasoning. The tasks span a wide range of historical problems—from factual retrieval based on primary sources to interpretive analysis of manuscripts and images, to interdisciplinary challenges involving archaeology, linguistics, or cultural history. Furthermore, the benchmark dataset spans 29 ancient and modern languages and covers a wide range of historical periods and world regions. Finding the poor performance of LLMs and other agents on HistBench, we further present HistAgent, a history-specific agent equipped with carefully designed tools for OCR, translation, archival search, and image understanding in History. On HistBench, HistAgent based on GPT-4o achieves an accuracy of 27.54% pass@1 and 36.47% pass@2, significantly outperforming LLMs with online search and generalist agents, including GPT-4o (18.60%), DeepSeek-R1(14.49%), Grok 3(17.63%) and Open Deep Research by smolagents(20.29% pass@1 and 25.12% pass@2). These results highlight the limitations of existing LLMs and generalist agents and demonstrate the advantages of HistAgent for historical reasoning. Notably, HistAgent also achieves 60.00% pass@1 accuracy on the GAIA benchmark, showing that domain-specific customization doesn't hinder HistAgent's competitive performance on real-world general tasks.

Deep Learning · Large Language Models

Nitay Calderon, Eyal Ben-David, Zorik Gekhman, Eran Ofek, Gal Yona

Standard factuality evaluations of LLMs treat all errors alike, obscuring whether failures arise from missing knowledge (empty shelves) or from limited access to encoded facts (lost keys). We propose a behavioral framework that profiles factual knowledge at the level of facts rather than questions, characterizing each fact by whether it is encoded, and then by how accessible it is: cannot be recalled, can be directly recalled, or can only be recalled with inference-time computation (thinking). To support such profiling, we introduce WikiProfile, a new benchmark constructed via an automated pipeline with a prompted LLM grounded in web search. Across 4 million responses from 13 LLMs, we find that encoding is nearly saturated in frontier models on our benchmark, with GPT-5 and Gemini-3 encoding 95--98\% of facts. However, recall remains a major bottleneck: many errors previously attributed to missing knowledge instead stem from failures to access it. These failures are systematic and disproportionately affect long-tail facts and reverse questions. Finally, we show that thinking improves recall and can recover a substantial fraction of failures, indicating that future gains may rely less on scaling and more on methods that improve how models utilize what they already encode.

Yifeng Gao, Yifan Ding, Li Wang, Feida Huang, Ye Sun, Yixu Wang, Xin Wang, Yutao Wu, Hanxun Huang, Yunhao Feng 等

The accelerating realism of AI-generated content has amplified the spread of deceptive information and eroded public trust. Prior works typically split the problem into two tracks, media authenticity, which concerns whether content is real or AI-generated, and content veracity, which concerns semantic and factual correctness, thereby missing their joint effects in practice. We present FakeWorld 1.0, which deeply fuses these two orthogonal axes into a unified omni-modal benchmark. Along the media axis, FakeWorld spans text, audio, image, and video synthesis; along the content axis, it instantiates cross-modal semantic inconsistencies and factual errors. These axes are jointly instantiated within realistic web-based and streaming-style presentation scenarios, reflecting how multimodal deception is composed and delivered in real-world settings. FakeWorld provides explainable annotations in the form of per-instance rationales, enabling transparent, evidence-based diagnosis. Under a unified protocol, our evaluation of open- and closed-source MLLMs exposes capacity limits and highlights FakeWorld’s effectiveness at surfacing mixed-source, high-fidelity deception. Beyond the benchmark, we introduce OmniCheck, a unified omni-modal agentic workflow that performs explainable detection across both axes and outputs evidence-backed reports. We aim for FakeWorld 1.0 to serve as a realistic stress test and a practical foundation for building future systems that enable scalable, explainable detection of fake multimodal content.

Théo Uscidda, Marta Gazulla, Maks Ovsjanikov, Federico Tombari, Leonidas Guibas

Current Large Reasoning Models (LRMs) exhibit remarkable general capabilities but significantly underperform in spatial reasoning tasks. Existing approaches treat this gap as a knowledge deficit, relying on supervised fine-tuning (SFT) to ingest labeled data from external vision sources or synthetic engines. In contrast, we argue that for many tasks, spatial capabilities are already present in pre-trained LRMs but require alignment through principles of internal logical coherence. In this work, we propose a self-supervised reinforcement learning (RL) framework that targets the internal Chain-of-Thought (CoT) process without requiring ground-truth annotations. By formalizing the notion of consistency verifiers—reward functions that check for geometric and semantic consistency under transformations like flipping or swapping the order of objects in the question—and optimizing them via our new OT-GRPO strategy, a minimal-consistency matching variant of group relative policy optimization, we demonstrate that models can self-correct their spatial logic. Our results show that this label-free consistency training approaches the accuracy of models trained with ground-truth supervision and achieves similar generalization across diverse tasks and domains.

Reinforcement Learning · Batch/Offline

Siyu Wang, Xiaocong Chen, Mingming Gong, Yong Li, Quan Sheng, Lina Yao

Transformer sequence models such as Decision Transformer can learn strong offline policies from logged trajectories, but they can suffer from causal confusion: reliance on spurious correlations that predict reward in the data but do not reflect the true causal mechanisms of the environment. We propose CSET (Counterfactual State Editing Transformer), which improves robustness in strictly offline reinforcement learning without learning environment transition dynamics. CSET first fits a causal reward model as a conditional variational autoencoder to infer a posterior over reward disturbances for each transition. Conditioning on the factual action and a sampled disturbance, a counterfactual state generator proposes a minimally edited state whose predicted reward matches the factual reward; a normalized move-band constraint and an acceptance gate enforce state plausibility and reward consistency. We then augment trajectories by replacing only the observation token with the edited state while keeping the next observation factual, so the policy is not trained on synthetic successor transitions. On the model side, CSET uses a causally structured hybrid transformer: separate convolutional encoders process return-to-go, state, and action streams for local temporal structure, and a final attention block is softly supervised so action prediction focuses on its direct causal parents (state and return-to-go). Experiments on D4RL locomotion and AntMaze tasks and on offline recommendation benchmarks show consistent gains over transformer baselines and substantially improved robustness to injected spurious distractors.

Social Aspects · Alignment

Yuanshuai Li, Yuping Yan, Junfeng Tang, Zeqi Zheng, Yaochu Jin

Multimodal Large Language Models (MLLMs) have significantly improved the performance of various tasks, but continue to suffer from visual hallucinations, a critical issue where generated responses contradict visual evidence. While Direct Preference Optimization (DPO) is widely used for alignment, its application to MLLMs often fails to capture fine-grained semantic differences and encourages shortcut learning. To address these challenges, we propose Semantic Curriculum Preference Optimization (SCPO), a novel framework for MLLM alignment. SCPO employs a progressive, easy-to-hard curriculum built upon our Semantic Curriculum Preference Pairs dataset, which provides fine-grained semantic contrasts sorted by difficulty. This curriculum is trained with a dynamic reference model and a novel symmetric, bidirectional objective to facilitate simultaneous learning from both textual and visual preferences. To our knowledge, SCPO is the first framework to unify semantics, symmetry, and curriculum for MLLMs alignment, effectively mitigating visual hallucinations. Extensive experiments on LLaVA models across various scales and versions validate that SCPO demonstrates superior performance compared to baseline models on multiple hallucination benchmarks, reducing the hallucination rate by up to 62.9%. Moreover, evaluations on generalized benchmarks show that SCPO improves factuality while preserving general capabilities, with its performance remaining stable across general vision-language benchmarks.

Deep Learning · Large Language Models

Xinyu Liu, Jun Wu

Large Language Models (LLMs) often produce hallucinated outputs, which limit their reliability in high-stakes applications. Conformal prediction can provide guarantees on the correctness and factuality of LLM outputs, but existing approaches rely on the exchangeability assumption, which rarely holds in online settings where user queries and interests change over time. To solve this problem, in this paper, we propose **PACE** (**P**roactive **A**daptive **C**onformal Inferenc**E**), a novel framework that sequentially updates the time-varying target miscoverage parameter with a dynamic step size to maintain valid coverage under online distribution shifts. PACE is motivated by the theoretical connections between expected miscoverage error and key factors such as distribution shifts and instantaneous parameter error. It integrates two complementary signals: (1) a proactive shift detection to estimate the magnitude of distribution shifts, and (2) a reactive error that scales updates according to the local coverage gap. Extensive experiments on synthetic and real-world datasets demonstrate that PACE consistently outperforms advanced adaptive baselines. It reduces the deviation from the target error rate by up to 60\% in QA tasks and accelerates coverage recovery by over 2.5x during abrupt shifts, ensuring stable factuality guarantees without compromising utility and stability.

Deep Learning · Large Language Models

Ting Wang, Yuanjie Shi, Yan Yan, Huan Zhang

Large language models (LLMs) increasingly perform multi-step reasoning, where intermediate claims form implicit directed acyclic graphs whose node correctness is structurally conditioned on their ancestors. This makes factuality uncertainty structural, rather than a trivial accumulation of node-wise errors, and necessitates inference-time uncertainty quantification over the reasoning structure. While conformal prediction (CP) offers flexible user-specified factuality control, existing work remains post-hoc and cannot intervene during generation. To fill the gap between CP’s flexibility and its post-hoc limitation, we propose an *Inference-Time Conformal Reasoning (ITCR)* framework that integrates CP directly into reasoning graph generation. ITCR learns a structure-level factuality uncertainty function that aggregates claim-level factuality signals over reasoning graphs without complex modeling assumptions. We then design the non-conformity score based on graph-level factuality uncertainty and calibrate the conformal threshold to decide when to stop generation. We theoretically show such generation is nested, yielding valid coverage guarantees for factuality control. Experiments over multiple datasets and coverage objectives demonstrate empirically valid coverage. In downstream reasoning tasks, inference-time calibrated graphs yield more accurate generation than post-hoc pruned graphs.