The paper presents C3T (Cross-modal Capabilities Conservation Test), a new benchmark for assessing the performance of speech-aware large language models. The benchmark utilizes textual tasks and a voice cloning text-to-speech model to quantify the extent to which language understanding capabilities are preserved when the model is accessed via speech input. C3T quantifies the fairness of the model for different categories of speakers and its robustness across text and speech modalities.

Efficient reasoning distillation for long chain-of-thought (CoT) models is increasingly constrained by the assumption of a single oracle teacher, despite the practical availability of multiple candidate teachers and growing CoT corpora. We revisit teacher selection and observe that different students have different "best teachers," and even for the same student, the best teacher can vary across datasets. Therefore, to unify multiple teachers' reasoning abilities into a student to overcome conflicts among various teachers' supervision, we propose Merge-of-Thought Distillation (MoT), a lightweight framework that alternates between teacher-specific supervised fine-tuning branches and weight-space merging of the resulting student variants. On competition math benchmarks, using only about 200 CoT samples, applying MoT to a Qwen3-14B student surpasses strong models including Deepseek-R1, Qwen3-32B, and OpenAI-O1, demonstrating substantial gains. Besides, MoT consistently outperforms the best single-teacher distillation, improves general reasoning beyond mathematics while reducing catastrophic forgetting, and shows robustness to distribution-shifted and peer-level teachers. Finally, we have demonstrated MoT possesses consensus CoT by eliminating teacher-specific inductive biases and inter-teacher conflicts while repeatedly reinforcing the learning of consensus reasoning features. These results position MoT as a simple, effective route to efficiently distilling long CoT capabilities from diverse teachers into compact students.

Reinforcement learning (RL) has proven effective in incentivizing the reasoning abilities of large language models (LLMs), but suffers from severe efficiency challenges due to its trial-and-error nature. While the common practice employs supervised fine-tuning (SFT) as a warm-up stage for RL, this decoupled two-stage approach suffers from catastrophic forgetting: second-stage RL gradually loses SFT-acquired behaviors and inefficiently explores new patterns. This study introduces a novel method for learning reasoning models that employs bilevel optimization to facilitate better cooperation between these training paradigms. By conditioning the SFT objective on the optimal RL policy, our approach enables SFT to meta-learn how to guide RL's optimization process. During training, the lower level performs RL updates while simultaneously receiving SFT supervision, and the upper level explicitly maximizes the cooperative gain-the performance advantage of joint SFT-RL training over RL alone. Empirical evaluations on five reasoning benchmarks demonstrate that our method consistently outperforms baselines and achieves a better balance between effectiveness and efficiency.

Abstract--Large language models (LLMs) have been widely applied to assist in finding solutions for diverse questions. Prior work has proposed representing a method as a pair of a question and its corresponding solution, enabling method reuse. However, existing approaches typically require the questions to be highly similar . In this paper, we extend the scope of method reuse to address questions with low similarity or with hidden similarities that are not explicitly observable. For questions that are similar in a general-specific sense (i.e., broader or narrower in scope), we propose to first separate the question and solution, rather than directly feeding the pair to the LLM. The LLM is then guided to adapt the solution to new but related questions, allowing it to focus on solution transfer rather than question recognition. Furthermore, we extend this approach to cases where questions only share partial features or hidden characteristics. This enables cross-question method reuse beyond conventional similarity constraints. Experimental verification shows that our scope-extension approach increases the probability of filtering out reusable solutions, thereby improving the effectiveness of cross-question method reuse. Large language models (LLMs) [1] are increasingly used to assist with complex tasks such as writing, reasoning [2], and problem solving [3]. Traditional LLMs are primarily trained at the word level to predict the next token given a context, or through masked word prediction by filling in missing tokens [4].

Retrieval-Augmented Generation (RAG) has advanced open-domain question answering by incorporating external information into model reasoning. However, effectively leveraging external information to enhance reasoning presents the following challenges: (1) low signal-to-noise ratio, where answer-supportive external information is diluted by irrelevant material, and (2) error accumulation, which arises in multi-hop reasoning when incomplete or misleading information is incorporated. To address these challenges, we introduce EviNote-RAG, a framework that follows a retrieve-note-answer workflow. Instead of reasoning directly over raw external information, the model first produces Supportive-Evidence Notes (SENs), which concisely preserve answer-critical information and explicitly mark key and uncertainty information to improve accuracy. We further design an entailment-based Evidence Quality Reward (EQR) to ensure that SENs are logically sufficient to derive the final answer, thereby enhancing SENs' quality. Experiments on both in-domain and out-of-domain QA benchmarks show that EviNote-RAG achieves state-of-the-art performance, improving answer accuracy, training stability, robustness, and efficiency. In particular, it yields relative F1 gains of 20% on HotpotQA (+0.093), 40% on Bamboogle (+0.151), and 91% on 2Wiki (+0.256), benefiting from improvements in the reasoning process.

We introduce PETLP (Privacy-by-design Extract, Transform, Load, and Present), a compliance framework that embeds legal safeguards directly into extended ETL pipelines. Central to PETLP is treating Data Protection Impact Assessments as living documents that evolve from preregistration through dissemination. Through systematic Red-dit analysis, we demonstrate how extraction rights fundamentally differ between qualifying research organisations (who can invoke DSM Article 3 to override platform restrictions) and commercial entities (bound by terms of service), whilst GDPR obligations apply universally. We demonstrate why true anonymisation remains unachievable for social media data and expose the legal gap between permitted dataset creation and uncertain model distribution. By structuring compliance decisions into practical workflows and simplifying institutional data management plans, PETLP enables researchers to navigate regulatory complexity with confidence, bridging the gap between legal requirements and research practice.

In this work, we present a simple yet theoretically motivated improvement to Supervised Fine-Tuning (SFT) for the Large Language Model (LLM), addressing its limited generalization compared to reinforcement learning (RL). Through mathematical analysis, we reveal that standard SFT gradients implicitly encode a problematic reward structure that may severely restrict the generalization capabilities of model. To rectify this, we propose Dynamic Fine-Tuning (DFT), stabilizing gradient updates for each token by dynamically rescaling the objective function with the probability of this token. With just a single-line change, the method outperforms standard SFT on multiple difficult benchmarks and base models, from math reasoning to code generation and multi-modal tasks, demonstrating improved generalization. Additionally, DFT achieves competitive results in offline RL settings, providing an effective yet streamlined alternative. Supervised Fine-Tuning (SFT), which adapts models to expert demonstrations, has become the standard post-training paradigm for Large Language Models (LLMs). It enables efficient task adaptation and capability enhancement (Chung et al., 2024; Zhang et al., 2024b; Sanh et al., 2022; Ouyang et al., 2022), and is popular for its ease of implementation and rapid acquisition of expert-like behaviors (Wei et al., 2022; Zhou et al., 2023). Despite these advantages, SFT often shows limited generalization compared to reinforcement learning (RL) (Chu et al., 2024; Ouyang et al., 2022; Christiano et al., 2017; Bai et al., 2022; Huan et al., 2025; Swamy et al., 2025). RL leverages explicit reward or verification signals to explore diverse strategies and thus generalizes better. However, RL requires substantial computation, careful hyperparameter tuning, and explicit reward signals--conditions often impractical in real-world settings (Schulman et al., 2017; Ouyang et al., 2022; Sheng et al., 2025; Strubell et al., 2019; Liu & Yin, 2024; Winsta, 2025). Moreover, RL can struggle to recover expert-like behaviors that SFT captures efficiently (Mandlekar et al., 2022; Chen et al., 2025b).

A large language model (LLM) with knowledge in both scientific and general tasks is the foundation of science general intelligence. However, directly continued pretraining an LLM using science data usually leads to catastrophic forgetting, which indicates severe degradation in general ability. In this report, we present Innovator, which solves this problem by upcycling a pre-trained dense LLM into a fine-grained Mixtures-of-Experts model during continued pretraining, where different experts are expected to learn science knowledge in different disciplines, and a shared expert is utilized for general tasks. Innovator introduces a four-stage upcycle training paradigm: (1) Scientific Expert Induction on discipline-specific data, (2) Fine-grained Expert Splitting via FFN dimension decomposition, (3) Science-Aware Routing warmup, and (4) Generalist-Scientist Integration training on hybrid datasets. Such a paradigm enables knowledge in the general domain, and different scientific disciplines can be decoupled, avoiding the negative influence among knowledge in different domains. With 53.3B total parameters and 13.3B activated, Innovator extends Qwen2.5-7B using a shared general expert and 64 specialized scientific experts with 8 activated. Trained on 300B tokens with tri-level quality-controlled data, Innovator achieves 25% average improvement across 30 scientific tasks with a win rate as 70%, while retaining 99% performance in general tasks. Furthermore, Innovator-Reason, which is post-trained from Innovator for reasoning boosting, exhibits excellent reasoning performance in solving complex scientific problems with improvements over 30%.

Deterministically controlling the target generation language of large multilingual language models (LLMs) remains a fundamental challenge, particularly in zero-shot settings where neither explicit language prompts nor fine-tuning are available. In this work, we investigate whether sparse autoencoder (SAE) features, previously shown to correlate with interpretable model behaviors, can be leveraged to steer the generated language of LLMs during inference. Leveraging pretrained SAEs on the residual streams of Gemma-2B and Gemma-9B, we identify features whose activations differ most significantly between English and four target languages: Chinese, Japanese, Spanish, and French. By modifying just a single SAE feature at one transformer layer, we achieve controlled language shifts with up to 90\% success, as measured by FastText language classification, while preserving semantic fidelity according to LaBSE (Language-Agnostic BERT Sentence Embedding) similarity. Our analysis reveals that language steering is most effective in mid-to-late transformer layers and is amplified by specific attention heads disproportionately associated with language-sensitive SAE features. These results demonstrate the promise of sparse feature steering as a lightweight and interpretable mechanism for controllable multilingual generation.

Objective Renal cancer is a common malignancy and a major cause of cancer-related deaths. Computed tomography (CT) is central to early detection, staging, and treatment planning. However, the growing CT workload increases radiologists' burden and risks incomplete documentation. Automatically generating accurate reports remains challenging because it requires integrating visual interpretation with clinical reasoning. Advances in artificial intelligence (AI), especially large language and vision-language models, offer potential to reduce workload and enhance diagnostic quality. Methods We propose a clinically informed, two-stage framework for automatic renal CT report generation. In Stage 1, a multi-task learning model detects structured clinical features from each 2D image. In Stage 2, a vision-language model generates free-text reports conditioned on the image and the detected features. To evaluate clinical fidelity, generated clinical features are extracted from the reports and compared with expert-annotated ground truth. Results Experiments on an expert-labeled dataset show that incorporating detected features improves both report quality and clinical accuracy. The model achieved an average AUC of 0.75 for key imaging features and a METEOR score of 0.33, demonstrating higher clinical consistency and fewer template-driven errors. Conclusion Linking structured feature detection with conditioned report generation provides a clinically grounded approach to integrate structured prediction and narrative drafting for renal CT reporting. This method enhances interpretability and clinical faithfulness, underscoring the value of domain-relevant evaluation metrics for medical AI development.