Reporting quality is an important topic in clinical trial research articles, as it can impact clinical decisions. In this article, we test the ability of large language models to assess the reporting quality of this type of article using the Consolidated Standards of Reporting Trials (CONSORT). We create CONSORT-QA, an evaluation corpus from two studies on abstract reporting quality with CONSORT-abstract standards. We then evaluate the ability of different large generative language models (from the general domain or adapted to the biomedical domain) to correctly assess CONSORT criteria with different known prompting methods, including Chain-of-thought. Our best combination of model and prompting method achieves 85% accuracy. Using Chain-of-thought adds valuable information on the model's reasoning for completing the task.

Safety-aligned Large Language Models (LLMs) still show two dominant failure modes: they are easily jailbroken, or they over-refuse harmless inputs that contain sensitive surface signals. We trace both to a common cause: current models reason weakly about links between actions and outcomes and over-rely on surface-form signals, lexical or stylistic cues that do not encode consequences. We define this failure mode as Consequence-blindness. To study consequence-blindness, we build a benchmark named CB-Bench covering four risk scenarios that vary whether semantic risk aligns with outcome risk, enabling evaluation under both matched and mismatched conditions which are often ignored by existing safety benchmarks. Mainstream models consistently fail to separate these risks and exhibit consequence-blindness, indicating that consequence-blindness is widespread and systematic. To mitigate consequence-blindness, we introduce CS-Chain-4k, a consequence-reasoning dataset for safety alignment. Models fine-tuned on CS-Chain-4k show clear gains against semantic-camouflage jailbreaks and reduce over-refusal on harmless inputs, while maintaining utility and generalization on other benchmarks. These results clarify the limits of current alignment, establish consequence-aware reasoning as a core alignment goal and provide a more practical and reproducible evaluation path.

Training fair and unbiased machine learning models is crucial for high-stakes applications, yet it presents significant challenges. Effective bias mitigation requires deep expertise in fairness definitions, metrics, data preprocessing, and machine learning techniques. In addition, the complex process of balancing model performance with fairness requirements while properly handling sensitive attributes makes fairness-aware model development inaccessible to many practitioners. To address these challenges, we introduce FairAgent, an LLM-powered automated system that significantly simplifies fairness-aware model development. FairAgent eliminates the need for deep technical expertise by automatically analyzing datasets for potential biases, handling data preprocessing and feature engineering, and implementing appropriate bias mitigation strategies based on user requirements. Our experiments demonstrate that FairAgent achieves significant performance improvements while significantly reducing development time and expertise requirements, making fairness-aware machine learning more accessible to practitioners.

Large language model multi-agent systems (LLM-MAS) offer a promising paradigm for harnessing collective intelligence to achieve more advanced forms of AI behaviour. While recent studies suggest that LLM-MAS can outperform LLM single-agent systems (LLM-SAS) on certain tasks, the lack of systematic experimental designs limits the strength and generality of these conclusions. We argue that a principled understanding of task complexity, such as the degree of sequential reasoning required and the breadth of capabilities involved, is essential for assessing the effectiveness of LLM-MAS in task solving. To this end, we propose a theoretical framework characterising tasks along two dimensions: depth, representing reasoning length, and width, representing capability diversity. We theoretically examine a representative class of LLM-MAS, namely the multi-agent debate system, and empirically evaluate its performance in both discriminative and generative tasks with varying depth and width. Theoretical and empirical results show that the benefit of LLM-MAS over LLM-SAS increases with both task depth and width, and the effect is more pronounced with respect to depth. This clarifies when LLM-MAS are beneficial and provides a principled foundation for designing future LLM-MAS methods and benchmarks.

Common evaluation paradigms for language models focus on scoring single responses through accuracy metrics or proper scoring rules, failing to capture the full richness of a model's belief state. Recent work illustrates that language models hallucinate in-part because they are optimised to be good test-takers under binary scoring schemes that reward any answer over abstention. While this insight naturally leads to penalty-based approaches, they ignore crucial distinctions in how models distribute uncertainty, for example between hedging toward incorrect answers versus hedging toward "I don't know" responses. A novel evaluation metric, the Distributional Correctness Score (DCS), is introduced to solve this problem, i.e., of not considering a model's entire probability distribution over answer choices. DCS naturally distinguishes between harmful overconfidence in wrong answers and uncertainty expressed through abstention, providing scores in an interpretable default range. Through theoretical analysis and illustrative examples, DCS is demonstrated to offer a more nuanced and aligned evaluation paradigm that incentivises models to express genuine uncertainty rather than guessing. Adapting 12 existing evaluation benchmarks to DCS's variants and measuring performance on six language models reveals that for half of the tested benchmarks scores are negative across all tested models, indicating significant tendencies towards hallucination. Evaluation of language models has commonly focused on whether they produce'correct' or desired outputs in response to given inputs or instructions, as measured using accuracy or probability-based scoring rules that account for confidence in model predictions. However, the paradigm of focusing on a single answer fundamentally misses a critical aspect of evaluating performance: how models distribute their beliefs across the space of possible responses, including the possibility of abstaining from answering in conditions of uncertainty. Recent work (Kalai et al., 2025) provides compelling evidence that language model'hallucinations' persist in-part due to the socio-technical problem of flawed evaluation metrics. Under traditional binary scoring - where correct answers receive a positive score (maximally 1 for perfect correctness), any response like "I don't know" (IDK) receives 0, and incorrect answers also receive 0 - the optimal strategy for any rational agent is to always guess rather than abstain, even when confidence in the guess is minimal. This creates a systematic bias in our evaluation paradigms toward overconfident responses and offers a socio-technical explanation for why language models persist in making confident assertions about uncertain information, i.e., 'hallucinate'.

Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning (PEFT) technique that adapts large pre-trained models by adding low-rank matrices to their weight updates. However, in the context of fine-tuning multi-head self-attention (MHA), LoRA has been employed to adapt each attention head separately, thereby overlooking potential synergies across different heads. To mitigate this issue, we propose a novel Hyper-shared Low-Rank Adaptation (HoRA) method, which utilizes joint hypernetworks to generate low-rank matrices across attention heads. By coupling their adaptation through a shared generator, HoRA encourages cross-head information sharing, and thus directly addresses the aforementioned limitation of LoRA. By comparing LoRA and HoRA through the lens of hierarchical mixture of experts, our theoretical findings reveal that the latter achieves superior sample efficiency to the former. Furthermore, through extensive experiments across diverse language and vision benchmarks, we demonstrate that HoRA outperforms LoRA and other PEFT methods while requiring only a marginal increase in the number of trainable parameters.

While large language models (LLMs) demonstrate impressive capabilities, their reliance on parametric knowledge often leads to factual inaccuracies. Retrieval-Augmented Generation (RAG) mitigates this by leveraging external documents, yet existing approaches treat retrieved passages as isolated chunks, ignoring valuable structure that is crucial for document organization. Motivated by this gap, we propose Retrieve-DocumentRoute-Read (RDR2), a novel framework that explicitly incorporates structural information throughout the RAG process. RDR2 employs an LLM-based router to dynamically navigate document structure trees, jointly evaluating content relevance and hierarchical relationships to assemble optimal evidence. Our key innovation lies in formulating document routing as a trainable task, with automatic action curation and structure-aware passage selection inspired by human reading strategies. Through comprehensive evaluation on five challenging datasets, RDR2 achieves state-of-the-art performance, demonstrating that explicit structural awareness significantly enhances RAG systems' ability to acquire and utilize knowledge, particularly in complex scenarios requiring multi-document synthesis.

The professionalism of a human doctor in outpatient service depends on two core abilities: the ability to make accurate medical decisions and the medical consultation skill to conduct strategic, empathetic patient inquiry. Existing Large Language Models (LLMs) have achieved remarkable accuracy on medical decision-making benchmarks. However, they often lack the ability to conduct the strategic and empathetic consultation, which is essential for real-world clinical scenarios. To address this gap, we propose Doctor-R1, an AI doctor agent trained to master both of the capabilities by ask high-yield questions and conduct strategic multi-turn inquiry to guide decision-making. Our framework introduces three key components: a multi-agent interactive environment, a two-tiered reward architecture that separately optimizes clinical decision-making and communicative inquiry skills, and an experience repository to ground policy learning in high-quality prior trajectories. We evaluate Doctor-R1 on OpenAI's HealthBench and MAQuE, assessed across multi-facet metrics, such as communication quality, user experience, and task accuracy. Remarkably, Doctor-R1 surpasses state-of-the-art open-source specialized LLMs by a substantial margin with higher parameter efficiency and outperforms powerful proprietary models. Furthermore, the human evaluations show a strong preference for Doctor-R1 to generate human-preferred clinical dialogue, demonstrating the effectiveness of the framework.

Multimodal large language models (MLLMs) hold promise for integrating diverse data modalities, but current medical adaptations such as LLaVA-Med often fail to fully exploit the synergy between color fundus photography (CFP) and optical coherence tomography (OCT), and offer limited interpretability of quantitative biomarkers. We introduce GROK, a grounded multimodal large language model that jointly processes CFP, OCT, and text to deliver clinician-grade diagnoses of ocular and systemic disease. GROK comprises three core modules: Knowledge-Guided Instruction Generation, CLIP-Style OCT-Biomarker Alignment, and Supervised Instruction Fine-Tuning, which together establish a quantitative-to-qualitative diagnostic chain of thought, mirroring real clinical reasoning when producing detailed lesion annotations. To evaluate our approach, we introduce the Grounded Ophthalmic Understanding benchmark, which covers six disease categories and three tasks: macro-level diagnostic classification, report generation quality, and fine-grained clinical assessment of the generated chain of thought. Experiments show that, with only LoRA (Low-Rank Adaptation) fine-tuning of a 7B-parameter Qwen2 backbone, GROK outperforms comparable 7B and 32B baselines on both report quality and fine-grained clinical metrics, and even exceeds OpenAI o3. Code and data are publicly available in the GROK repository.

Children learn to speak with a low amount of data and can be taught new words on a few-shot basis, making them particularly data-efficient learners. The BabyLM challenge aims at exploring language model (LM) training in the low-data regime but uses metrics that concentrate on the head of the word distribution. Here, we introduce LongTail-Swap (LT-Swap), a benchmark that focuses on the tail of the distribution, i.e., measures the ability of LMs to learn new words with very little exposure, like infants do. LT-Swap is a pretraining corpus-specific test set of acceptable versus unacceptable sentence pairs that isolate semantic and syntactic usage of rare words. Models are evaluated in a zero-shot fashion by computing the average log probabilities over the two members of each pair. We built two such test sets associated with the 10M words and 100M words BabyLM training sets, respectively, and evaluated 16 models from the BabyLM leaderboard. Our results not only highlight the poor performance of language models on rare words but also reveal that performance differences across LM architectures are much more pronounced in the long tail than in the head. This offers new insights into which architectures are better at handling rare word generalization. We've also made the code publicly avail