Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2609.00206v1 Announce Type: cross Abstract: Despite their growing use in video moderation, multimodal large language models (MLLMs) exhibit a compositional safety blind spot: videos composed of...
arXiv:2609.00866v1 Announce Type: cross Abstract: The rapid advancement of vision-language models (VLMs) has accelerated progress in computational pathology; however, whole-slide image (WSI)-based pa...
arXiv:2609.00999v1 Announce Type: cross Abstract: When a question has valid answers under different normative frameworks, a language model must decide which framework to use and whether it can answer...
arXiv:2609.01046v1 Announce Type: cross Abstract: Production LLMs must handle inputs that attempt to override system instructions, bypass safety policies or elicit harmful responses. A common mitigat...
arXiv:2609.01120v1 Announce Type: cross Abstract: Early recognition of lane-change intention is essential for proactive decision-making in autonomous driving and advanced driver assistance systems. T...
arXiv:2609.01210v1 Announce Type: cross Abstract: Safety benchmarks for large language models often assess the risk of a user query, although the outcome of question answering depends on whether the...
arXiv:2609.01418v1 Announce Type: cross Abstract: To mitigate the sample complexity of real-world reinforcement learning (RL), a common practice is to first train a policy in a simulator, where sampl...
arXiv:2609.01487v1 Announce Type: cross Abstract: Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable chan...
arXiv:2609.01491v1 Announce Type: cross Abstract: The growing rate at which LLM agents interact with one another raises key questions about language evolution in multi-LLM-agent settings, with implic...
arXiv:2609.01556v1 Announce Type: cross Abstract: We evaluate embedding retrieval where surface form and meaning are pulled apart on purpose: retrieving items that share underlying structure but not...
arXiv:2603.10359v2 Announce Type: replace Abstract: Distilling reasoning capabilities from Large Reasoning Models (LRMs) into smaller models is typically constrained by the limitations of rejection s...
arXiv:2609.00213v1 Announce Type: new Abstract: Reinforcement learning fine-tuning of large language models increasingly adopts multiple reward dimensions, including verifiable rules, task-specific e...
arXiv:2609.01548v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly used in advice seeking and decision making that may affect social judgements. Despite stigma's profound e...
arXiv:2609.00565v1 Announce Type: cross Abstract: Cultural fine-tuning has become the de facto paradigm for building culture-aware large language models (LLMs), yet existing optimization exclusively...
arXiv:2609.01147v1 Announce Type: cross Abstract: Text-rich visual inputs require models that can read, retrieve, and compress language directly in pixel space, yet existing pixel-text encoders strug...
arXiv:2609.01432v1 Announce Type: cross Abstract: Scientific citations carry rhetorical intent. Scholars may cite prior work positively (supporting), negatively (contrasting), or neutrally (mentionin...
arXiv:2609.01456v1 Announce Type: cross Abstract: Composed image retrieval (CIR) retrieves a target image from a reference image and a text modification. This paper studies metadata-available CIR rer...
arXiv:2609.00909v1 Announce Type: new Abstract: Reliable evaluation of automated coronary computed tomography angiography (CCTA) report generation requires standardized multicentre benchmarks and cli...
arXiv:2609.01433v1 Announce Type: new Abstract: Concept erasure aims to suppress unsafe, privacy-sensitive, or undesirable generations in text-to-image diffusion models while preserving benign semant...
JanusMesh introduces a fast, training‑free framework for creating 3D visual illusion meshes that reveal different semantics from various viewpoints. The method splits generation into two stages: a cross‑space dual‑branch denoising process that aligns 3D latents with CLIP guidance and blends Signed Distance Fields for seamless geometry, followed by a view‑conditioned texture synthesis module that aggregates 2D diffusion priors onto the fused mesh. Experiments show that JanusMesh produces highly realistic, dual‑semantic 3D illustrations in only 3–5 minutes, outperforming prior approaches in geometric integrity, semantic recognizability, and efficiency.