Open-ended medical LLM evaluation remains weakly grounded in physician-calibrated coverage of clinically relevant response criteria, especially in localized clinical settings. We introduce ClinConsensus, a Chinese medical benchmark of 2,500 expert-curated cases spanning 36 specialties, 12 task themes, multiple difficulty levels, and lay-facing versus professional-facing settings. Each case is paired with 30 case-specific binary rubric criteria. To evaluate whether responses satisfy enough physician-authored criteria, we propose Clinician-Anchored Coverage Score (CACS), a physician-calibrated threshold metric instantiated at k = 10, and develop a dual-judge framework combining a GPT-5.1 grader with a physician-supervised Qwen3-8B judge. Evaluating 11 frontier LLMs, we find a persistent coverage gap: Rubric Accuracy ranges from 39.6% to 52.1%, whereas CACS@10 ranges from 17.8% to 32.9%, leaving a 19.2–21.9 point gap across models. Stratified analyses further reveal substantial variation across reasoning, evidence use, structured extraction, medication instructions, follow-up, and dialogue register. These results suggest that medical LLM evaluation should measure thresholded, rubric-grounded clinical coverage rather than average partial correctness.

Large Language Models (LLMs) excel at programming tasks but frequently fail at deterministic, fine-grained reasoning in natural language, relying heavily on semantic approximations rather than robust symbolic execution. To bridge this gap, we propose MIMIC, a framework that leverages executable code as a rigorous medium for reasoning data synthesis. MIMIC fundamentally transforms algorithms into verifiable reasoning trajectories through narrative fusion, code-guided test synthesis, and dynamic code instrumentation. Crucially, these explicit intermediate execution states naturally form a Code-Instrumented Reward (CIR), providing dense, high-fidelity process supervision for reinforcement learning without external reward models. Extensive evaluations reveal that models trained via SFT and GRPO on our synthesized dataset achieve substantial, consistent gains. Our method significantly elevates accuracy across general reasoning, complex mathematical benchmarks, and fine-grained deterministic tasks, demonstrating that the procedural rigor of executable code can effectively unlock and enhance the generalized reasoning capabilities of LLMs. Our code and data are available at https://github.com/zjy1298/MIMIC.

Autoregressive generation in large language models (LLMs) conventionally decodes from the final layer, assuming that deeper representations yield more reliable next-token predictions. We revisit this assumption by revealing a recurring **Guess–Refine–Perturb** dynamic: early layers form coarse guesses, intermediate layers refine reasoning-relevant semantics, and final layers can perturb these refined predictions toward generic or alignment-preferred tokens. We introduce **Confident Decoding**, a training-free decoding strategy that dynamically selects a near-final layer through entropy-guided conservative backward search. We further formulate layer selection as an optimal stopping problem and show that, under explicit local-monotonicity and bounded-probe-noise conditions, the search rule filters terminal perturbations while bounding unfinished refinement. Experiments across dense and Mixture-of-Experts LLMs show that gains concentrate on challenging reasoning settings, including GPQA-Diamond, Omni-MATH, and HLE, while saturated tasks remain broadly stable, with no additional KV-cache memory and less than 2% latency increase. These results suggest dynamically bypassing final-layer perturbations can recover stronger reasoning behavior from aligned LLMs.

As large language models (LLMs) are increasingly applied to legal domain-specific tasks, evaluating their ability to perform legal work in real-world settings has become essential. However, existing legal benchmarks rely on simplified and highly standardized tasks, failing to capture the ambiguity, complexity, and reasoning demands of real legal practice. Moreover, prior evaluations often adopt coarse, single-dimensional metrics and do not explicitly assess fine-grained legal reasoning. To address these limitations, we introduce PLawBench, a Practical Law Benchmark designed to evaluate LLMs in realistic legal practice scenarios. Grounded in real-world legal workflows, PLawBench models the core processes of legal practitioners through three task categories: public legal consultation, practical case analysis, and legal document generation. These tasks assess a model’s ability to identify legal issues and key facts, perform structured legal reasoning, and generate legally coherent documents. PLawBench comprises 850 questions across 13 practical legal scenarios, with each question accompanied by expert-designed evaluation rubrics, resulting in approximately 12,500 rubric items for fine-grained assessment. Using an LLM-based evaluator aligned with human expert judgments, we evaluate 10 state-of-the-art LLMs. Experimental results show that none achieves strong performance on PLawBench, revealing substantial limitations in the fine-grained legal reasoning capabilities of current LLMs and highlighting important directions for future evaluation and development of legal LLMs. Data is available at: https://anonymous.4open.science/r/PLawbench-B524/.

Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize Outcome Accuracy, which undermines their ability to generalize during RLHF. We introduce Rationale Consistency, a fine-grained metric that quantifies the alignment between the model’s reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

Recent advancements in Large Language Models (LLMs) have shown outstanding potential for role-playing applications. Evaluating these capabilities is becoming crucial yet remains challenging. Existing benchmarks mostly adopt a character-centric approach, simplify user-character interactions to isolated Q&A tasks, and fail to reflect real-world applications. To address this limitation, we introduce RMTBench, a comprehensive user-centric bilingual role-playing benchmark featuring 80 diverse characters and over 8,000 dialogue rounds. RMTBench includes custom characters with detailed backgrounds and abstract characters defined by simple traits, enabling evaluation across various user scenarios. Our benchmark constructs dialogues based on explicit user motivations rather than character descriptions, ensuring alignment with practical user applications. Furthermore, we construct an authentic multi-turn dialogue simulation mechanism. With carefully selected evaluation dimensions and LLM-based scoring, this mechanism captures the complex intention of conversations between the user and the character. By shifting focus from character background to user intention fulfillment, RMTBench bridges the gap between academic evaluation and practical deployment requirements, offering a more effective framework for assessing role-playing capabilities in LLMs. All code and datasets will be released soon.

In this paper, we propose a novel language model guided captioning approach, LAMOC, for knowledge-based visual question answering (VQA). Our approach employs the generated captions by a captioning model as the context of an answer prediction model, which is a Pre-Trained Language model (PLM). As the major contribution, we leverage the guidance and feedback of the prediction model to improve the capability of the captioning model. In this way, the captioning model can become aware of the task goal and information need from the PLM. To develop our approach, we design two specific training stages, where the first stage adapts the captioning model to the prediction model (selecting more suitable caption propositions for training) and the second stage tunes the captioning model according to the task goal (learning from feedback of the PLM). Extensive experiments demonstrate the effectiveness of the proposed approach on the knowledge-based VQA task. Specifically, on the challenging A-OKVQA dataset, LAMOC outperforms several competitive zero-shot methods and even achieves comparable results to a fine-tuned VLP model. Our code is publicly available at https://github.com/RUCAIBox/LAMOC.

Pre-trained language models (PLMs) have achieved remarkable success in natural language generation (NLG) tasks. Up to now, most NLG-oriented PLMs are pre-trained in an unsupervised manner using the large-scale general corpus. In the meanwhile, an increasing number of models pre-trained with labeled data (i.e. “supervised pre-training”) showcase superior performance compared to unsupervised pre-trained models. Motivated by the success of supervised pre-training, we propose Multi-task superVised Pre-training (MVP) for natural language generation. We collect a large-scale natural language generation corpus, MVPCorpus, from 77 datasets over 11 diverse NLG tasks. Then we unify these examples into a general text-to-text format to pre-train the text generation model MVP in a supervised manner. For each task, we further pre-train specific soft prompts to stimulate the model’s capacity to perform a specific task. Our MVP model can be seen as a practice that utilizes recent instruction tuning on relatively small PLMs. Extensive experiments have demonstrated the effectiveness and generality of our MVP model in a number of NLG tasks, which achieves state-of-the-art performance on 13 out of 17 datasets, outperforming BART by 9.3% and Flan-T5 by 5.8%.

Pretrained language models (PLMs) encode a large amount of world knowledge. However, as such knowledge is frozen at the time of model training, the models become static and limited by the training data at that time. In order to further improve the capacity of PLMs for knowledge-intensive tasks, we consider augmenting PLMs with the large-scale web using search engine. Unlike previous augmentation sources (e.g., Wikipedia data dump), the web provides broader, more comprehensive and constantly updated information. In this paper, we present a web-augmented PLM – UniWeb, which is trained over 16 knowledge-intensive tasks in a unified text-to-text format. Instead of simply using the retrieved contents from web, our approach has made two major improvements. Firstly, we propose an adaptive search engine assisted learning method that can self-evaluate the confidence level of PLM’s predictions, and adaptively determine when to refer to the web for more data, which can avoid useless or noisy augmentation from web. Secondly, we design a pretraining task, i.e., continual knowledge learning, based on salient spans prediction, to reduce the discrepancy between the encoded and retrieved knowledge. Experiments on a wide range of knowledge-intensive tasks show that our model significantly outperforms previous retrieval-augmented methods.

People often imagine relevant scenes to aid in the writing process. In this work, we aim to utilize visual information for composition in the same manner as humans. We propose a method, LIVE, that makes pre-trained language models (PLMs) Learn to Imagine for Visually-augmented natural language gEneration. First, we imagine the scene based on the text: we use a diffusion model to synthesize high-quality images conditioned on the input texts. Second, we use CLIP to determine whether the text can evoke the imagination in a posterior way. Finally, our imagination is dynamic, and we conduct synthesis for each sentence rather than generate only one image for an entire paragraph. Technically, we propose a novel plug-and-play fusion layer to obtain visually-augmented representations for each text. Our vision-text fusion layer is compatible with Transformer-based architecture. We have conducted extensive experiments on four generation tasks using BART and T5, and the automatic results and human evaluation demonstrate the effectiveness of our proposed method. We will release the code, model, and data at the link: https://github.com/RUCAIBox/LIVE.

Nowadays, pretrained language models (PLMs) have dominated the majority of NLP tasks. While, little research has been conducted on systematically evaluating the language abilities of PLMs. In this paper, we present a large-scale empirical study on general language ability evaluation of PLMs (ElitePLM). In our study, we design four evaluation dimensions, memory, comprehension, reasoning, and composition, to measure ten widely-used PLMs within five categories. Our empirical results demonstrate that: (1) PLMs with varying training objectives and strategies are good at different ability tests; (2) fine-tuning PLMs in downstream tasks is usually sensitive to the data size and distribution; (3) PLMs have excellent transferability between similar tasks. Moreover, the prediction results of PLMs in our experiments are released as an open resource for more deep and detailed analysis on the language abilities of PLMs. This paper can guide the future work to select, apply, and design PLMs for specific tasks. We have made all the details of experiments publicly available at https://github.com/RUCAIBox/ElitePLM.

Pretrained language models (PLMs) have made remarkable progress in text generation tasks via fine-tuning. While, it is challenging to fine-tune PLMs in a data-scarce situation. Therefore, it is non-trivial to develop a general and lightweight model that can adapt to various text generation tasks based on PLMs. To fulfill this purpose, the recent prompt-based learning offers a potential solution. In this paper, we improve this technique and propose a novel prompt-based method (PTG) for text generation in a transferable setting. First, PTG learns a set of source prompts for various source generation tasks and then transfers these prompts as target prompts to perform target generation tasks. To consider both task- and instance-level information, we design an adaptive attention mechanism to derive the target prompts. For each data instance, PTG learns a specific target prompt by attending to highly relevant source prompts. In extensive experiments, PTG yields competitive or better results than fine-tuning methods. We release our source prompts as an open resource, where users can add or reuse them to improve new text generation tasks for future research. Code and data can be available at https://github.com/RUCAIBox/Transfer-Prompts-for-Text-Generation.