Software development relies heavily on extensive unit testing, which makes the efficiency of automated Unit Test Generation (UTG) particularly important. However, most existing LLMs generate test cases one token at a time in each forward pass, which leads to inefficient UTG. Recently, diffusion LLMs (dLLMs) have emerged, offering promising parallel generation capabilities and showing strong potential for efficient UTG. Despite this advantage, their application to UTG is still constrained by a clear trade-off between efficiency and test quality, since increasing the number of tokens generated in each step often causes a sharp decline in the quality of test cases. To overcome this limitation, we present DiffTester, an acceleration framework specifically tailored for dLLMs in UTG. The key idea of DiffTester is that unit tests targeting the same focal method often share repetitive structural patterns. By dynamically identifying these common patterns through abstract syntax tree analysis during generation, DiffTester adaptively increases the number of tokens produced at each step without compromising the quality of the output. To enable comprehensive evaluation, we extend the original TestEval benchmark, which was limited to Python, by introducing additional programming languages including Java and C++. Extensive experiments on three benchmarks with two representative models show that DiffTester delivers significant acceleration while preserving test coverage. Moreover, DiffTester generalizes well across different dLLMs and programming languages, providing a practical and scalable solution for efficient UTG in software development. Code and data are publicly available at https://github.com/wellbeingyang/DLM4UTG-open .

Strategy Choice Theory (SCT; Siegler and Shrager, 1984; Siegler, 2000) explains important aspects of children's arithmetic learning based upon principles including learning from developmentally naturalistic data, probabilistic representation, confidence-based retrieval, and the phase-like importance of scaffolding strategies, such as finger-counting. Here we recast SCT as a ``Small Math Model'' (SMM), employing a neural-network-based architecture analogous to LLMs. The SMM extends SCT to include counting practice, symbol (number) embedding, and gated attention. Similar to earlier work, the SMM demonstrates constructive and destructive interference between counting and addition, and the ``wave-like'' use of finger-counting as sum recall improves. We plan to extend the SMM to later aspects of the decades-long SCT program, including adaptive strategy choice and eventually strategy discovery, providing a unified platform to investigate the understanding of numerical characteristics and relationships essential for mathematical reasoning -- as it can emerge in LLM-based agents.

Training large language models with Reinforcement Learning with Verifiable Rewards (RLVR) exhibits a set of distinctive and puzzling behaviors that remain poorly understood, including a two-stage learning curve, a V-shaped response-length trajectory, and a pronounced vulnerability to catastrophic forgetting. In this work, we propose that these behaviors are emergent collective phenomena governed not by neural implementation details, but by the topological evolution of the latent reasoning graph in semantic space. By demonstrating a dynamical isomorphism between a 1.5B-parameter LLM and a minimal Concept Network Model (CoNet), we trace the causal source to the self-organization of a sparse concept web pinned to an average degree of two. This geometric perspective provides a unified physical explanation for the observed anomalies: the V-shaped trajectory tracks the evolution from parallel local skill optimization to global network integration; catastrophic forgetting stems from the topological disconnection of critical ``trunk'' edges; and policy collapse arises from the accumulation of sequential transitions at the web's leaf nodes, where broad exploration abruptly freezes into rigid, high-reward trajectories. Identifying a ``maximally frustrated state'' at the transition between learning stages, we propose Annealed-RLVR, a principled algorithm that injects a targeted SFT ``heating'' step to resolve this topological bottleneck. Experiments confirm that this theory-driven intervention outperforms standard RLVR on both in-distribution and out-of-distribution benchmarks (including Minerva and AIME). By recasting RLVR from black-box optimization into a predictable process of structural self-organization, our work provides a new physical intuition for engineering the emergent reasoning capabilities of future AI systems.

When AI interacts with the physical world -- as a robot or an assistive agent -- new safety challenges emerge beyond those of purely ``digital AI". In such interactions, the potential for physical harm is direct and immediate. How well do state-of-the-art foundation models understand common-sense facts about physical safety, e.g. that a box may be too heavy to lift, or that a hot cup of coffee should not be handed to a child? In this paper, our contributions are three-fold: first, we develop a highly scalable approach to continuous physical safety benchmarking of Embodied AI systems, grounded in real-world injury narratives and operational safety constraints. To probe multi-modal safety understanding, we turn these narratives and constraints into photorealistic images and videos capturing transitions from safe to unsafe states, using advanced generative models. Secondly, we comprehensively analyze the ability of major foundation models to perceive risks, reason about safety, and trigger interventions; this yields multi-faceted insights into their deployment readiness for safety-critical agentic applications. Finally, we develop a post-training paradigm to teach models to explicitly reason about embodiment-specific safety constraints provided through system instructions. The resulting models generate thinking traces that make safety reasoning interpretable and transparent, achieving state of the art performance in constraint satisfaction evaluations. The benchmark is released at https://asimov-benchmark.github.io/v2

How do Large Language Models understand moral dimensions compared to humans? This first large-scale Bayesian evaluation of market-leading language models provides the answer. In contrast to prior work using deterministic ground truth (majority or inclusion rules), we model annotator disagreements to capture both aleatoric uncertainty (inherent human disagreement) and epistemic uncertainty (model domain sensitivity). We evaluated the best language models (Claude Sonnet 4, DeepSeek-V3, Llama 4 Maverick) across 250K+ annotations from nearly 700 annotators in 100K+ texts spanning social networks, news and forums. Our GPU-optimized Bayesian framework processed 1M+ model queries, revealing that AI models typically rank among the top 25\% of human annotators, performing much better than average balanced accuracy. Importantly, we find that AI produces far fewer false negatives than humans, highlighting their more sensitive moral detection capabilities.

Activation steering offers a promising approach to controlling the behavior of Large Language Models by directly manipulating their internal activations. However, most existing methods struggle to jointly steer multiple attributes, often resulting in interference and undesirable trade-offs. To address this challenge, we propose Multi-Subspace Representation Steering (MSRS), a novel framework for effective multi-attribute steering via subspace representation fine-tuning. MSRS reduces inter-attribute interference by allocating orthogonal subspaces to each attribute, isolating their influence within the model's representation space. MSRS also incorporates a hybrid subspace composition strategy: it combines attribute-specific subspaces for unique steering directions with a shared subspace for common steering directions. A dynamic weighting function learns to efficiently integrate these components for precise control. During inference, MSRS introduces a token-level steering mechanism that dynamically identifies and intervenes on the most semantically relevant tokens, enabling fine-grained behavioral modulation. Experimental results show that MSRS significantly reduces attribute conflicts, surpasses existing methods across a range of attributes, and generalizes effectively to diverse downstream tasks. These models often exhibit undesirable behaviors, including toxicity, bias, or factual inaccuracies, rooted in the complex and opaque representations learned during training (Y ang et al., 2024c; Wang et al., 2025a; Zhang et al., 2025b). Effectively controlling these behaviors without compromising model performance remains an open research problem (Jiao et al., 2025). Recently, activation steering methods offer a promising avenue for behavior adjustment by manipulating model activations post-training (Im & Li, 2025). Compared to fine-tuning, they offer lightweight control without the need for retraining or access to model weights, enabling scalable adaptation to diverse downstream tasks. These approaches derive an activation steering vector from the difference between the activations of positive and negative samples, applying it during inference to guide outputs toward desired properties without altering model parameters (Rimsky et al., 1 The left output contains biases and falsehoods. MSRS reduces attribute conflicts by separating steering spaces and using a shared subspace to capture common properties, enabling better integration.

Multimodal large language models (MLLMs) demonstrate exceptional performance in vision-language tasks, yet their processing of long videos is constrained by input context length and high computational costs. Sparse frame sampling thus becomes a necessary preprocessing step, with sampled frame quality directly impacting downstream performance. Existing keyframe search algorithms achieve a balance between efficiency and sampled frame quality but heavily rely on the visual modality alone. This makes them difficult to adapt to text-related tasks and often leads to retrieval results deviating from core semantic content. To address this, we propose the VISUAL-SUBTITLE INTEGRATION (VSI), a multimodal keyframe retrieval framework. It employs a dual-branch collaborative retrieval approach combining Video Search and Subtitle Match to fuse complementary visual and textual information for precise localization. Experiments on LongVideoBench and VideoMME demonstrate that VSI achieves state-of-the-art accuracy in keyframe retrieval while delivering breakthrough performance in text-related tasks and exhibiting strong generalization across other tasks.

The integration of Large Language Models (LLMs) into Virtual Reality (VR) games marks a paradigm shift in the design of immersive, adaptive, and intelligent digital experiences. This paper presents a comprehensive review of recent research at the intersection of LLMs and VR, examining how these models are transforming narrative generation, non-player character (NPC) interactions, accessibility, personalization, and game mastering. Drawing from an analysis of 62 peer reviewed studies published between 2018 and 2025, we identify key application domains ranging from emotionally intelligent NPCs and procedurally generated storytelling to AI-driven adaptive systems and inclusive gameplay interfaces. We also address the major challenges facing this convergence, including real-time performance constraints, memory limitations, ethical risks, and scalability barriers. Our findings highlight that while LLMs significantly enhance realism, creativity, and user engagement in VR environments, their effective deployment requires robust design strategies that integrate multimodal interaction, hybrid AI architectures, and ethical safeguards. The paper concludes by outlining future research directions in multimodal AI, affective computing, reinforcement learning, and open-source development, aiming to guide the responsible advancement of intelligent and inclusive VR systems.

In the past few years LLMs have emerged as a tool that can aid programmers by taking natural language descriptions and generating code based on it. However, the reliability of LLM code generation and current validation techniques for it are far from strong enough to be used for mission-critical or safety-critical applications. In this work we explore ways to offer formal guarantees of correctness to LLM generated code; such guarantees could improve the quality of general AI Code Assistants and support their use for critical applications. To address this challenge we propose to incorporate a Formal Query Language that can represent a user's intent in a formally defined but natural language-like manner that a user can confirm matches their intent. We then have a formal specification of the user intent which we can use to verify that LLM-generated code matches the user's intent. We implement these ideas in our system, Astrogator, for the Ansible programming language, widely used for system administration, including for critical systems. The system includes an intuitive formal query language, a calculus for representing the behavior of Ansible programs, and a symbolic interpreter and a unification algorithm which together are used for the verification. A key innovation in Astrogator is the use of a Knowledge Base to capture system-specific implementation dependencies that greatly reduce the need for system knowledge in expressing formal queries. On a benchmark suite of 21 code-generation tasks, our verifier is able to verify correct code in 83% of cases and identify incorrect code in 92%.

We examined how online hate is motivated by receiving social approval via Walther's (2024) social approval theory of online hate, which argues (H1a) more signals of social approval on hate messages predicts more subsequent hate messages, and (H1b) as social approval increases, hate speech becomes more extreme. Using 110 million messages from Parler (2018-2021), we observed the number of upvotes received on a hate speech post was unassociated with hate speech in one's next post and during the next month, three-months, and six-months. The number of upvotes received on (extreme) hate speech comments, however, was positively associated with (extreme) hate speech during the next week, month, three-months, and six-months. Between-person effects revealed an average positive relationship between social approval and hate speech production at all time intervals. For comments, social approval linked more strongly to online hate than social disapproval. Social approval is a critical mechanism facilitating online hate propagation.