Existing LLM-based role-playing methods often rely on superficial textual descriptions or simplistic metrics, inadequately modeling both intrinsic and extrinsic character dimensions. Additionally, they typically simulate character memory with implicit model knowledge or basic retrieval augment generation without explicit memory alignment, compromising memory consistency. The two issues weaken reliability of role-playing LLMs in several applications, such as trustworthy social simulation. To address these limitations, we propose PsyMem, a novel framework integrating fine-grained psychological attributes and explicit memory control for role-playing. PsyMem supplements textual descriptions with 26 psychological indicators to detailed model character. Additionally, PsyMem implements memory alignment training, explicitly trains the model to align character's response with memory, thereby enabling dynamic memory-controlled responding during inference. By training Qwen2.5-7B-Instruct on our specially designed dataset (including 5,414 characters and 38,962 dialogues extracted from novels), the resulting model, termed as PsyMem-Qwen, outperforms baseline models in role-playing, achieving the best performance in human-likeness and character fidelity.

LLM-based formal proof assistants (e.g., in Lean) hold great promise for automating mathematical discovery. But beyond syntactic correctness, do these systems truly understand mathematical structure as humans do? We investigate this question in context of mathematical inequalities -- specifically the prover's ability to recognize that the given problem simplifies by applying a known inequality such as AM/GM. Specifically, we are interested in their ability to do this in a compositional setting where multiple inequalities must be applied as part of a solution. We introduce Ineq-Comp, a benchmark built from elementary inequalities through systematic transformations, including variable duplication, algebraic rewriting, and multi-step composition. Although these problems remain easy for humans, we find that most provers -- including Goedel, STP, and Kimina-7B -- struggle significantly. DeepSeek-Prover-V2-7B shows relative robustness, but still suffers a 20% performance drop (pass@32). Even for DeepSeek-Prover-V2-671B model, the gap between compositional variants and seed problems exists, implying that simply scaling up the model size alone does not fully solve the compositional weakness. Strikingly, performance remains poor for all models even when formal proofs of the constituent parts are provided in context, revealing that the source of weakness is indeed in compositional reasoning. Our results expose a persisting gap between the generalization behavior of current AI provers and human mathematical intuition. All data and evaluation code can be found at https://github.com/haoyuzhao123/LeanIneqComp.

Survival analysis often relies on Cox models, assuming both linearity and proportional hazards (PH). This study evaluates machine and deep learning methods that relax these constraints, comparing their performance with penalized Cox models on a benchmark of three synthetic and three real datasets. In total, eight different models were tested, including six non-linear models of which four were also non-PH. Although Cox regression often yielded satisfactory performance, we showed the conditions under which machine and deep learning models can perform better. Indeed, the performance of these methods has often been underestimated due to the improper use of Harrell's concordance index (C-index) instead of more appropriate scores such as Antolini's concordance index, which generalizes C-index in cases where the PH assumption does not hold. In addition, since occasionally high C-index models happen to be badly calibrated, combining Antolini's C-index with Brier's score is useful to assess the overall performance of a survival method. Results on our benchmark data showed that survival prediction should be approached by testing different methods to select the most appropriate one according to sample size, non-linearity and non-PH conditions. To allow an easy reproducibility of these tests on our benchmark data, code and documentation are freely available at https://github.com/compbiomed-unito/survhive.

Abstract-- Large language models (LLMs) have demonstrated strong capabilities in code generation, underscoring the critical need for rigorous and comprehensive evaluation. Existing evaluation approaches fall into three categories, including human-centered, metric-based, and LLM-based. Considering that human-centered approaches are labour-intensive and metric-based ones overly rely on reference answers, LLM-based approaches are gaining increasing attention due to their stronger contextual understanding capabilities. However, they generally evaluate the generated code based on static prompts, and tend to fail for complex code scenarios which typically involve multiple requirements and require more contextual information. In addition, these approaches lack fine-grained evaluation for complex code, resulting in limited explainability. T o mitigate the limitations, we propose CodeVisionary, the first agent-based evaluation framework for complex code generation. CodeVisionary consists of two stages: (1) Requirement-guided multi-dimensional context distillation stage, which first formulates a detailed evaluation plan by decomposing task requirements, and then stepwise collects multi-dimensional contextual information for each requirement. A comprehensive evaluation report is also generated for enhanced explainability. For validation, we construct a new benchmark consisting of 363 samples spanning 37 coding scenarios and 23 programming languages. Extensive experiments demonstrate that CodeVisionary achieves the best performance among three baselines for evaluating complex code generation, outperforming the best baseline with average improvements of 0.217, 0.163, and 0.141 in Pearson, Spearman, and Kendall-T au coefficients, respectively. With the rapid development of large language models (LLMs), these models have demonstrated promising results in code generation [1], [2].

Epilepsy is typically diagnosed through electroencephalography (EEG) and long-term video-EEG (vEEG) monitoring. The manual analysis of vEEG recordings is time-consuming, necessitating automated tools for seizure detection. Recent advancements in machine learning have shown promise in real-time seizure detection and prediction using EEG and video data. However, diversity of seizure symptoms, markup ambiguities, and limited availability of multimodal datasets hinder progress. This paper reviews the latest developments in automated video-EEG analysis and discusses the integration of multimodal data. We also propose a novel pipeline for treatment effect estimation from vEEG data using concept-based learning, offering a pathway for future research in this domain.

Even when prompts and parameters are secured, transformer language models remain vulnerable because their key-value (KV) cache during inference constitutes an overlooked attack surface. This paper introduces Malicious Token Injection (MTI), a modular framework that systematically perturbs cached key vectors at selected layers and timesteps through controlled magnitude and frequency, using additive Gaussian noise, zeroing, and orthogonal rotations. A theoretical analysis quantifies how these perturbations propagate through attention, linking logit deviations to the Frobenius norm of corruption and softmax Lipschitz dynamics. Empirical results show that MTI significantly alters next-token distributions and downstream task performance across GPT-2 and LLaMA-2/7B, as well as destabilizes retrieval-augmented and agentic reasoning pipelines. These findings identify cache integrity as a critical yet underexplored vulnerability in current LLM deployments, positioning cache corruption as a reproducible and theoretically grounded threat model for future robustness and security research.

While reasoning-based large language models excel at complex tasks through an internal, structured thinking process, a concerning phenomenon has emerged that such a thinking process can aggregate social stereotypes, leading to biased outcomes. However, the underlying behaviours of these language models in social bias scenarios remain underexplored. In this work, we systematically investigate mechanisms within the thinking process behind this phenomenon and uncover two failure patterns that drive social bias aggregation: 1) stereotype repetition, where the model relies on social stereotypes as its primary justification, and 2) irrelevant information injection, where it fabricates or introduces new details to support a biased narrative. Building on these insights, we introduce a lightweight prompt-based mitigation approach that queries the model to review its own initial reasoning against these specific failure patterns. Experiments on question answering (BBQ and StereoSet) and open-ended (BOLD) benchmarks show that our approach effectively reduces bias while maintaining or improving accuracy.

While advancements in quantization have significantly reduced the computational costs of inference in deep learning, training still predominantly relies on complex floating-point arithmetic. Low-precision fixed-point training presents a compelling alternative. This work introduces a novel enhancement in low-precision logarithmic fixed-point training, geared towards future hardware accelerator designs. We propose incorporating bitwidth in the design of approximations to arithmetic operations. To this end, we introduce a new hardware-friendly, piece-wise linear approximation for logarithmic addition. Using simulated annealing, we optimize this approximation at different precision levels. A C++ bit-true simulation demonstrates training of VGG-11 and VGG-16 models on CIFAR-100 and TinyImageNet, respectively, using 12-bit integer arithmetic with minimal accuracy degradation compared to 32-bit floating-point training. Our hardware study reveals up to 32.5% reduction in area and 53.5% reduction in energy consumption for the proposed LNS multiply-accumulate units compared to that of linear fixed-point equivalents.

The use of reinforcement learning (RL) with chain-of-thought (CoT) reasoning has emerged as a promising approach for developing more capable language models. In turn, this has led to investigation of CoT monitoring as a compelling method for detecting harmful behaviors such as reward hacking, under the assumption that models' reasoning processes reflect their internal decision-making. In practice, LLM training often produces unintended behaviors due to imperfect reward signals, leading models to develop misaligned tendencies. A common corrective approach is to apply post-hoc instructions to avoid problematic behaviors like sycophancy, but what happens to the model's reasoning process when these instructions conflict with learned behaviors? We investigate this question in simple settings and find that models engage in systematic motivated reasoning--generating plausible-sounding justifications for violating their instructions while downplaying potential harms. Beyond being an interesting property of training, we find that while motivated reasoning can be detected by most frontier reasoning models, smaller LLM judges can fail to identify a portion of it, and in rare cases can themselves be persuaded that the reasoning is correct, despite it contradicting clear instructions. This capability gap raises concerns that as models become more sophisticated, their motivated reasoning may become increasingly difficult for monitors to detect. Our results underscore the need to account for motivated reasoning when relying on chain-of-thought processes for model evaluation and oversight. Figure 1: We perform RL finetuning on Llama 3 8B Instruct on behaviors of different kinds. When asked to act against their trained behaviors in evaluations throughout training, models transition from performing mostly genuine reasoning to highly motivated reasoning, twisting the constitutional principles provided to them in the prompt to support the behaviors incentivized via training. The integration of reinforcement learning (RL) and chain-of-thought (CoT) reasoning has emerged as a promising approach for developing more capable language models (Jaech et al., 2024; Guo et al., 2025). Recent work has shown that encouraging models to output "thinking tokens" before committing to a final answer leads to impressive performance, especially on tasks with verifiable answers where rewards can be automatically generated, such as mathematics and programming problems (Shao et al., 2024; Zhu et al., 2024).

Tool-augmented large language models (LLMs) are increasingly employed in real-world applications, but tool usage errors still hinder their reliability. We introduce ToolCritic, a diagnostic framework that evaluates and improves LLM behavior in multi-turn, tool-augmented dialogues. ToolCritic detects eight distinct error types specific to tool-calling (e.g., premature invocation, argument misalignment, and misinterpretation of tool outputs) and provides targeted feedback to the main LLM. The main LLM, assumed to have strong reasoning, task understanding and orchestration capabilities, then revises its response based on ToolCritic's feedback. We systematically define these error categories and construct a synthetic dataset to train ToolCritic. Experimental results on the Schema-Guided Dialogue (SGD) dataset demonstrate that ToolCritic improves tool-calling accuracy by up to 13% over baselines, including zero-shot prompting and self-correction techniques. This represents a promising step toward more robust LLM integration with external tools in real-world dialogue applications.