In real clinical practice, clinicians must sift through noisy and often conflicting information, progressively gathering and sequencing evidence before reaching conclusions. However, existing tuning methods for medical AI models are typically monologue-based - that is, models are fine-tuned on static question-answering (QA) tasks or medical articles, which fail to reflect the interactive and iterative nature of clinical reasoning. To bridge this gap, we introduce MuddyMaze, a benchmark designed to expose the limitations of current monologue-based tuning, and construct a large dialogue dataset of 22.2k doctor-patient interactions that capture stepwise diagnostic reasoning validated by medical experts. Building on those, we propose dialogue-tuning, a new fine-tuning paradigm that captures the internal reasoning dynamics unfolding across interactions. To assess the effectiveness of our approach, we evaluated dialogue-tuned models on MuddyMaze, where they outperform monologue-tuned baselines (e.g., MedQA) by +16.1% in one-round and +4.1% in multi-round evidence ranking, while maintaining or even improving accuracy on standard medical QA benchmarks (e.g., PubMedQA). These results indicate that dialogue-tuning not only enhances reasoning robustness and evidence integration but also preserves the factual precision of traditional QA performance.

Reasoning-capable large language models (LLMs) achieve strong performance on complex tasks but often exhibit overthinking after distillation, generating unnecessarily long chain-of-thought (CoT) reasoning even for simple inputs and incurring high inference cost. However, naively shortening reasoning length can degrade reasoning accuracy, as concise reasoning may be insufficient for certain inputs and lacks explicit supervision. We propose Auto Long-Short Reasoning (AutoL2S), a distillation framework that empowers non-reasoning LLMs to think thoroughly but only when necessary. AutoL2S first learns a lightweight switching token with verified long-short CoTs to enable instance-wise long-short reasoning selection. Then it leverages long-short reasoning rollouts induced by a switching token in a GRPO-style loss to improve reasoning efficiency while maintaining accuracy. Experiments demonstrate that AutoL2S effectively reduces reasoning length up to 71

Large Language Models (LLMs) suffer inference-time memory bottlenecks dominated by the attention Key-Value (KV) cache, which scales with model size and context length. While KV-cache quantization alleviates this cost, bit allocation between keys and values is often tuned heuristically, lacking theoretical grounding and generalizability. This paper proposes two theorems that anchor mixed-precision KV quantization in the intrinsic geometry of Transformer models. First, key projections systematically have larger spectral and Frobenius norms than value matrices, implying higher information density along the key path. Second, for any given memory budget, prioritizing precision for keys over values strictly reduces quantization error and better preserves accuracy. Empirical evaluations across various prominent LLMs and benchmarks show that key-favored allocations (e.g., 4-bit keys, 2-bit values) retain up to 98.3% accuracy compared to uniform allocations (e.g., 4-bit for both), while conserving memory. These results transform bit allocation from ad hoc tuning into a theoretically grounded, geometry-driven design principle for efficient LLM inference. Source code is available at https://github.com/mohsenhariri/spectral-kv.

Advances in Large Language Models (LLMs) have paved the way for Multimodal Large Language Models (MLLMs). Among these, Large Audio Language Models (LALMs) are essential for realizing universal auditory intelligence. Despite their remarkable performance, the escalation of LALMs' capabilities has significantly outpaced the development of systemic frameworks to ensure their trustworthiness. This survey provides a comprehensive investigation into the endogenous mechanisms of LALMs, detailing the architectural innovations and alignment algorithms that facilitate emergent reasoning. Specifically, we analyze how the transition to unified end-to-end frameworks and the integration of continuous acoustic signals expand the attack surface. To rigorously evaluate the risks within these paradigms, we establish a comprehensive taxonomy of trustworthiness, categorizing critical vulnerabilities such as cross-modal jailbreaking, latent acoustic backdoors, and biometric privacy leakage. We review the state-of-the-art LALMs through six analytical pillars: hallucination, robustness, safety, privacy, fairness, and authentication. The pronounced imbalance between a mature offensive landscape and underdeveloped defenses highlights persistent trustworthiness gaps and multidimensional risks in audio-centric intelligence. Finally, we propose a roadmap advocating for “Defense-in-Depth” architectures, causal auditory world modeling, and intrinsic representation engineering to support the development of more reliable and trustworthy audio intelligence. Our project has been uploaded to GitHub https://github.com/Kwwwww74/Awesome-Trustworthy-AudioLLMs.

Automated methods for the assessment of replicability of scientific claims offer a scalable complement to replication studies and traditional peer review. Drawing on a large dataset of claims, human judgments, and a limited set of replication outcomes, we developed and evaluated three distinct artificial intelligence systems designed to predict human expert assessments of replicability using diverse methodologies—including synthetic prediction markets, interpretable feature-based modeling, knowledge graph reasoning, and semantic parsing with argument structures. While these systems achieved modest calibration to human judgment distributions, they failed to discriminate between replicable and non-replicable claims. Our findings suggest that while machine assessments of research replicability may complement human reasoning, their current performance limitations and opportunities for bias demand careful evaluation before real-world application.

In the field of crisis/disaster informatics, social media is increasingly being used for improving situational awareness to inform response and relief efforts. Efficient and accurate text classification tools have been a focal area of investigation in crisis informatics. However, current methods mostly rely on single-label text classification models, which fails to capture different insights embedded in dynamic and multifaceted disaster-related social media data. This study introduces a novel approach to disaster text classification by enhancing a pre-trained Large Language Model (LLM) through instruction fine-tuning targeted for multi-label classification of disaster-related tweets. Our methodology involves creating a comprehensive instruction dataset from disaster-related tweets, which is then used to fine-tune an open-source LLM, thereby embedding it with disaster-specific knowledge. This fine-tuned model can classify multiple aspects of disaster-related information simultaneously, such as the type of event, informativeness, and involvement of human aid, significantly improving the utility of social media data for situational awareness in disasters. The results demonstrate that this approach enhances the categorization of critical information from social media posts, thereby facilitating a more effective deployment for situational awareness during emergencies. This research paves the way for more advanced, adaptable, and robust disaster management tools, leveraging the capabilities of LLMs to improve real-time situational awareness and response strategies in disaster scenarios.

Modern open-world agents such as OpenClaw exhibit powerful cross-environment execution capabilities yet introduce broad new safety risk sources. Meanwhile, advanced frontier AI models drastically lower attack barriers, rendering current agent alignment frameworks inadequate for real-world deployment. To tackle these emerging threats, we propose a lightweight and scalable agent safety alignment framework. Specifically, we update the agent safety taxonomy to accommodate emergent risks from Codex and OpenClaw execution scenarios. We further build a taxonomy-guided data engine with influence-function purification to train lightweight AgentDoG 1.5 variants (0.8B, 2B, 4B, and 8B parameters) using only around 1k samples, achieving comparable performance with leading closed-source models (e.g., GPT-5.4). Based on AgentDoG 1.5, we construct a highly efficient agentic safety SFT and RL training environment, which reduces deployment overhead in Docker-level environments by two orders of magnitude. Finally, we deploy AgentDoG 1.5 as a training-free online guardrail for real-time safety moderation. Extensive experimental results indicate that AgentDoG 1.5 achieves state-of-the-art performance in diverse and complex interactive agentic scenarios. All models and datasets are openly released.

Large Language Models (LLMs) have become proficient in addressing complex tasks by leveraging their extensive internal knowledge and reasoning capabilities. However, the black-box nature of these models complicates the task of explaining their decision-making processes. While recent advancements demonstrate the potential of leveraging LLMs to self-explain their predictions through natural language (NL) explanations, their explanations may not accurately reflect the LLMs' decision-making process due to a lack of fidelity optimization on the derived explanations. Measuring the fidelity of NL explanations is a challenging issue, as it is difficult to manipulate the input context to mask the semantics of these explanations. To this end, we introduce FaithLM to explain the decision of LLMs with NL explanations. Specifically, FaithLM designs a method for evaluating the fidelity of NL explanations by incorporating the contrary explanations to the query process. Moreover, FaithLM conducts an iterative process to improve the fidelity of derived explanations. Experiment results on three datasets from multiple domains demonstrate that FaithLM can significantly improve the fidelity of derived explanations, which also provides a better alignment with the ground-truth explanations.

As Graph Neural Networks (GNNs) have been widely used in real-world applications, model explanations are required not only by users but also by legal regulations. However, simultaneously achieving high fidelity and low computational costs in generating explanations has been a challenge for current methods. In this work, we propose a framework of GNN explanation named L e A rn R emoval-based A ttribution (LARA) to address this problem. Specifically, we introduce removal-based attribution and demonstrate its substantiated link to interpretability fidelity theoretically and experimentally. The explainer in LARA learns to generate removal-based attribution which enables providing explanations with high fidelity. A strategy of subgraph sampling is designed in LARA to improve the scalability of the training process. In the deployment, LARA can efficiently generate the explanation through a feed-forward pass. We benchmark our approach with other state-of-the-art GNN explanation methods on six datasets. Results highlight the effectiveness of our framework regarding both efficiency and fidelity. In particular, LARA is 3.1 \(\times\) faster and achieves higher fidelity than the state-of-the-art method on the large dataset ogbn-arxiv (more than 160K nodes and 1M edges), showing its great potential in real-world applications. Our source code is available at https://github.com/yaorong0921/LARA .

Current discussion surrounding the clinical capabilities of generative language models (GLMs) predominantly center around multiple-choice question-answer (MCQA) benchmarks derived from clinical licensing examinations. While accepted for human examinees, characteristics unique to GLMs bring into question the validity of such benchmarks. Here, we validate four benchmarks using eight GLMs, ablating for parameter size and reasoning capabilities, validating via prompt permutation three key assumptions that underpin the generalizability of MCQA-based assessments: that knowledge is applied, not memorized, that semantic consistency will lead to consistent answers, and that situations with no answers can be recognized. While large models are more resilient to our perturbations compared to small models, we globally invalidate these assumptions, with implications for reasoning models. Additionally, despite retaining the knowledge, small models are prone to answer memorization. All models exhibit significant failure in null-answer scenarios. We then suggest several adaptations for more robust benchmark designs more reflective of real-world conditions.

Vision-Language Models (VLMs) demand substantial computational resources during inference, largely due to the extensive visual input tokens for representing visual information. Previous studies have noted that visual tokens tend to receive less attention than text tokens, suggesting their lower importance during inference and potential for pruning. However, their methods encounter several challenges: reliance on greedy heuristic criteria for token importance and incompatibility with FlashAttention and KV cache. To address these issues, we introduce TopV, a compatible TOken Pruning with inference Time Optimization for fast and low-memory VLM, achieving efficient pruning without additional training or fine-tuning. Instead of relying on attention scores, we formulate token pruning as an optimization problem, accurately identifying important visual tokens while remaining compatible with FlashAttention. Additionally, since we only perform this pruning once during the prefilling stage, it effectively reduces KV cache size. Our optimization framework incorporates a visual-aware cost function considering factors such as Feature Similarity, Relative Spatial Distance, and Absolute Central Distance, to measure the importance of each source visual token, enabling effective pruning of low-importance tokens. Extensive experiments demonstrate that our method outperforms previous token pruning methods, validating the effectiveness and efficiency of our approach.

Liver transplantation is a critical treatment for end-stage liver diseases, but ensuring fair and effective allocation of scarce donor organs remains a major challenge in healthcare. Existing allocation policies often struggle to balance clinical outcomes with fairness across patients and demographic groups. To address this, we propose FairAlloc, a learning-based framework that formulates organ allocation as a ranking problem and jointly optimizes for post-transplant outcomes and fairness. Specifically, FairAlloc integrates two fairness objectives—group fairness across sensitive attributes such as gender and race and individual fairness between similar patients—into the ranking process. We evaluate FairAlloc using real-world data from the Organ Procurement and Transplantation Network (OPTN) and compare its performance to six baseline methods. Results show that FairAlloc improves group and individual fairness metrics by up to 37.9

Large language models (LLMs) are increasingly deployed and democratized on edge devices. To improve the efficiency of on-device deployment, small language models (SLMs) are often adopted due to their efficient decoding latency and reduced energy consumption. However, these SLMs often generate inaccurate responses when handling complex queries. One promising solution is uncertainty-based SLM routing, offloading high-stakes queries to stronger LLMs when resulting in low-confidence responses on SLM. This follows the principle of "If you lack confidence, seek stronger support" to enhance reliability. Relying on more powerful LLMs is yet effective but increases invocation costs. Therefore, striking a routing balance between efficiency and efficacy remains a critical challenge. Additionally, efficiently generalizing the routing strategy to new datasets remains under-explored. In this paper, we conduct a comprehensive investigation into benchmarking and generalization of uncertainty-driven routing strategies from SLMs to LLMs over 1500+ settings. Our findings highlight: First, uncertainty-correctness alignment in different uncertainty quantification (UQ) methods significantly impacts routing performance. Second, uncertainty distributions depend more on both the specific SLM and the chosen UQ method, rather than downstream data. Building on the insight, we propose a calibration data construction instruction pipeline and open-source a constructed hold-out set to enhance routing generalization on new downstream scenarios. The experimental results indicate calibration data effectively bootstraps routing performance without any new data.

Although Large Language Models (LLMs) perform well in general fields, they exhibit a confidence distortion problem on multi-choice question-answering (MCQA), particularly as the number of answer choices increases. Specifically, on MCQA with many choices, LLMs suffer from under-confidence in correct predictions and over-confidence in incorrect ones, leading to a substantially degraded performance. To solve this problem, we propose Self-ensemble in this work. Our method splits the choices into several groups and ensembles LLM predictions across these groups to reach a final decision. The advantage of Self-ensemble is its plug-and-play nature, where it can be integrated into existing LLM architecture based on a designed attention mask and positional encoding, without requiring labeled datasets for parameter tuning. Experimental results on three LLMs and datasets demonstrate that Self-ensemble comprehensively addresses the confidence distortion problem of LLMs, outperforming standard inference as well as baseline methods.

Large Language Models (LLMs) have grown rapidly in size, creating significant challenges for efficient deployment on resource-constrained hardware. In this paper, we introduce Dynamic-Length Float (DFloat11), a lossless compression framework that reduces LLM size by 30% while preserving outputs that are bit-for-bit identical to the original model. DFloat11 is motivated by the low entropy in the BFloat16 weight representation of LLMs, which reveals significant inefficiency in existing storage format. By applying entropy coding, DFloat11 assigns dynamic-length encodings to weights based on frequency, achieving near information-optimal compression without any loss of precision. To facilitate efficient inference with dynamic-length encodings, we develop a custom GPU kernel for fast online decompression. Our design incorporates the following: (i) decomposition of memory-intensive lookup tables (LUTs) into compact LUTs that fit in GPU SRAM, (ii) a two-phase kernel for coordinating thread read/write positions using lightweight auxiliary variables, and (iii) transformer-block-level decompression to minimize latency. Experiments on recent models, including Llama-3.1, Qwen-2.5, and Gemma-3, validates our hypothesis that DFloat11 achieves around 30% model size reduction while preserving bit-for-bit exact outputs. Compared to a potential alternative of offloading parts of an uncompressed model to the CPU to meet memory constraints, DFloat11 achieves 1.9-38.8x higher throughput in token generation. With a fixed GPU memory budget, DFloat11 enables 5.3-13.17x longer context lengths than uncompressed models. Notably, our method enables lossless inference of Llama-3.1-405B, an 810GB model, on a single node equipped with 8x80GB GPUs. Our code and models are available at this https URL

Single-cell genomics has revolutionized our understanding of cellular heterogeneity, but automating its analysis remains an open challenge. Cell-type annotation represents a critical bottleneck, particularly as datasets grow in size and complexity. While foundation models have shown promise in addressing this challenge, existing approaches require extensive fine-tuning for effective cell-type annotation. Here, we present Lemur, a single-cell foundation model specifically designed for Drosophila melanogaster . Lemur achieves fine-tuning-free cell-type annotation through comprehensive pre-training on an integrated whole-organism atlas with a unified cell-type annotation schema. To leverage this unified schema, we developed a dedicated hierarchical cell-type decoder architecture. This approach enables Lemur to generate consistent cell-type predictions across multiple levels of granularity without requiring additional training on new datasets. The model demonstrates strong performance across diverse tissue types, experimental conditions, and sequencing technologies. It also achieves batch-effect correction without explicitly training for this task. This automated analysis capability positions Lemur as an effective tool for the fly research community. Beyond its immediate applications, Lemur establishes a framework for accelerating biological discovery. It enables rapid iteration between computational predictions and experimental validation in the highly controlled Drosophila system, with potential implications for translational research in human biology, particularly in aging and neurodegenerative disease studies. ### Competing Interest Statement The authors have declared no competing interest.

Large Language Models (LLMs) have grown rapidly in size, creating significant challenges for efficient deployment on resource-constrained hardware. In this paper, we introduce Dynamic-Length Float (DFloat11), a lossless compression framework that reduces LLM size by 30 bit-for-bit identical to the original model. DFloat11 is motivated by the low entropy in the BFloat16 weight representation of LLMs, which reveals significant inefficiency in existing storage format. By applying entropy coding, DFloat11 assigns dynamic-length encodings to weights based on frequency, achieving near information-optimal compression without any loss of precision. To facilitate efficient inference with dynamic-length encodings, we develop a custom GPU kernel for fast online decompression. Our design incorporates the following: (i) decomposition of memory-intensive lookup tables (LUTs) into compact LUTs that fit in GPU SRAM, (ii) a two-phase kernel for coordinating thread read/write positions using lightweight auxiliary variables, and (iii) transformer-block-level decompression to minimize latency. Experiments on recent models, including Llama-3.1, Qwen-2.5, and Gemma-3, validates our hypothesis that DFloat11 achieves around 30 while preserving bit-for-bit exact outputs. Compared to a potential alternative of offloading parts of an uncompressed model to the CPU to meet memory constraints, DFloat11 achieves 1.9-38.8x higher throughput in token generation. With a fixed GPU memory budget, DFloat11 enables 5.3-13.17x longer context lengths than uncompressed models. Notably, our method enables lossless inference of Llama-3.1-405B, an 810GB model, on a single node equipped with 8x80GB GPUs. Our code and models are available at https://github.com/LeanModels/DFloat11.

Large Language Models (LLMs) exhibit exceptional proficiency in comprehending human language. Despite their significant success across a wide array of tasks, understanding tabular data remains a challenging task. Especially, tabular data lacks an intrinsic order of the different features (table fields), whereas LLMs take only sequential inputs. Consequently, an artificial order is imposed, the impact of which on the performance of LLMs has not yet been thoroughly investigated. Surprisingly, as discovered in this work, this artificially induced order bias dramatically influences the performance of LLMs on tasks related to tabular data. Mitigating the order bias presents a significant challenge. To address this, we propose a simple and cost-effective method, Re-Ordering Tabular feATures fOR LLM (ROTATOR-LLM), to conduct test-time compute without fine-tuning the base LLM. Aiming at optimizing the feature order of tabular data and boosting LLMs' capability to better understand the data semantics, ROTATOR-LLM re-frames the ordering problem as a feature trajectory generation task. A dynamic programming based meta-controller is trained to auto-regressively generate an individualized feature trajectory for each data instance via accumulative value estimation of the serialized feature input through the LLM's final performance metrics. Model performance is maximized by iteratively selecting features across different steps. Experimental results on multiple datasets and LLMs show close to or over 20% performance boosts via features reordered by ROTATOR-LLM against the un-ordered counterpart. Meanwhile, it outperforms stateof- the-Art tabular LLM methods with significant margin.

Anomaly detection (AD) is an important machine learning task with many real-world uses, including fraud detection, medical diagnosis, and industrial monitoring. Within natural language processing (NLP), AD helps detect issues like spam, misinformation, and unusual user activity. Although large language models (LLMs) have had a strong impact on tasks such as text generation and summarization, their potential in AD has not been studied enough. This paper introduces AD-LLM, the first benchmark that evaluates how LLMs can help with NLP anomaly detection. We examine three key tasks: (i) zero-shot detection, using LLMs' pre-trained knowledge to perform AD without tasks-specific training; (ii) data augmentation, generating synthetic data and category descriptions to improve AD models; and (iii) model selection, using LLMs to suggest unsupervised AD models. Through experiments with different datasets, we find that LLMs can work well in zero-shot AD, that carefully designed augmentation methods are useful, and that explaining model selection for specific datasets remains challenging. Based on these results, we outline six future research directions on LLMs for AD.

Large Language Models (LLMs) are now integral across various domains and have demonstrated impressive performance. Progress, however, rests on the premise that benchmark scores are both accurate and reproducible. We demonstrate that the reproducibility of LLM performance is fragile: changing system configuration, such as evaluation batch size, GPU count, and GPU version, can introduce significant differences in the generated responses. This issue is especially pronounced in reasoning models, where minor rounding differences in early tokens can cascade into divergent chains of thought, ultimately affecting accuracy. For instance, under bfloat16 precision with greedy decoding, a reasoning model like DeepSeek-R1-Distill-Qwen-7B can exhibit up to 9\% variation in accuracy and 9,000 tokens difference in response length due to differences in GPU count, type, and evaluation batch size. We trace the root cause of this variability to the non-associative nature of floating-point arithmetic under limited numerical precision. This work presents the first systematic investigation into how numerical precision affects reproducibility in LLM inference. Through carefully controlled experiments across various hardware, software, and precision settings, we quantify when and how model outputs diverge. Our analysis reveals that floating-point precision—while critical for reproducibility—is often neglected in evaluation practices. Inspired by this, we develop a lightweight inference pipeline, dubbed LayerCast, that stores weights in 16-bit precision but performs all computations in FP32, balancing memory efficiency with numerical stability. Code is available at https://github.com/nanomaoli/llm_reproducibility.