A new benchmark, NOHARM, evaluates 20 large language models (LLMs) and 4 clinical AI tools on 1,100 medical consultation cases, finding that direct use of AI-generated recommendations could result in severe harm in up to 24.6% of cases, with omission errors accounting for over 80% of severe errors. In a randomized study of 101 physicians, AI assistance improved performance, but physicians often omitted valuable AI recommendations, indicating complementary strengths in human-AI teaming.
Why it matters: This study provides the first systematic measurement of clinical safety in LLM-generated medical advice, revealing that widely used AI tools can produce potentially harmful recommendations and highlighting the need for explicit safety evaluation.
Research→Official→arXiv Audio and Speech Processing
A new preprint demonstrates that audio-language embedding models, such as CLAP, fail to distinguish between affirmative and negated sound concepts, mapping both to nearly identical representations. The authors introduce NegEval-Audio, a framework that converts existing datasets into negation-aware tasks, revealing that model performance drops sharply when handling negation—even for recent multimodal LLM-based models. While a training-free steering method offers some improvement for one task, the overall limitation persists.
Why it matters: This work exposes a fundamental limitation in current audio-language models, showing that standard evaluation overlooks their inability to handle negation, which may impact real-world applications requiring nuanced understanding.
SlimPer introduces a new approach to personalized ranking in recommendation systems by iteratively refining a compact knowledge base, rather than relying on large intermediate representations that scale with user history length. This design achieves O(N) per-layer computational cost and fixed-size intermediate representations, allowing the model to handle over 10,000 user history events efficiently. Deployed on Instagram Reels and Feed, SlimPer has demonstrated measurable improvements in user engagement while unifying sparse, dense, and sequence features within a single model backbone.
Why it matters: SlimPer enables deeper and more efficient personalization in large-scale recommendation systems by decoupling model depth from user history length, reducing compute and memory requirements.
Researchers introduce EG-VAR, a Lean 4-based tool-calling architecture that ensures every verified output is grounded in attested tool calls and kernel-checked inference, thereby eliminating unsupported claims. On a subset of TableBench numerical reasoning tasks (n=120), EG-VAR achieves perfect accuracy (120/120) compared to a 95% baseline, and maintains 100% source-faithfulness on stress tests where baselines drop to 80-90%. The system also provides explicit audit trails for abstentions and formalization errors.
Why it matters: EG-VAR offers a practical and auditable approach to eliminating LLM hallucination in empirical inference, potentially transforming trustworthiness in high-stakes AI applications.
MESH is a unified retrieval scaling framework designed to address the scaling bias of heterogeneity in large-scale recommendation systems. Through a modularized architecture with gated bias correction, MESH partitions the feature space to reduce interference between different content types, leading to a 14x improvement in the power-law scaling exponent for fresh items. In online evaluations on Pinterest's Related Pins platform, MESH demonstrated a +5.5% lift in fresh-item repins, a 55% improvement in funnel efficiency, and a +0.46% improvement in user retention.
Why it matters: MESH represents a significant advance in unifying fragmented retrieval systems, improving scalability and performance for diverse content in large-scale recommendation platforms.
Researchers introduce MobileSAM2, a lightweight version of the SAM2 segmentation model designed for mobile devices. The model leverages a novel Hypergraphical Knowledge Distillation (HyperKD) technique, which includes Temporal and Granularity HyperKD to transfer temporal and multi-granularity knowledge from the original SAM2. MobileSAM2 is developed through architecture search and demonstrates promising results on segmentation tasks, including embodied AI benchmarks.
Why it matters: MobileSAM2 enables advanced segmentation capabilities on resource-constrained devices, expanding the practical deployment of spatial intelligence applications.
A new preprint audits data leakage in whole-slide image (WSI) multimodal benchmarks and finds that patient-level and institutional-level overlaps significantly compromise reported zero-shot performance of vision-language models. The authors document case-level train-test overlaps of 92.3–100% on TCGA-derived benchmarks and show that this leakage is linearly decodable from foundation-model features. They recommend concrete steps for contamination-free evaluation practices.
Why it matters: This work reveals that widely-used WSI VQA benchmarks may not accurately measure genuine multimodal reasoning, calling into question reported advances in computational pathology.
Researchers introduce ACID, a training-free wrapper that adaptively adjusts caching thresholds during video diffusion model inference. By dynamically switching between low and high thresholds based on the rate of change in the drift signal, ACID achieves up to 2.16x speedup over no-caching baselines and up to 38% additional speedup over conservative fixed-threshold caching, with negligible quality degradation (<0.3 dB PSNR). ACID is compatible with existing caching methods such as TeaCache, EasyCache, and DiCache, and works across multiple video diffusion models.
Why it matters: This approach advances the speed-quality tradeoff in video diffusion models, making them more practical for real-time applications without retraining.
A new framework called DM-KG introduces a structured knowledge graph to enhance spatial reasoning in vision-language models (VLMs) applied to street view imagery. By extracting and encoding directional and metric relationships between entities from 2D images, DM-KG significantly reduces distance estimation and direction judgment errors on spatial question-answering benchmarks.
Why it matters: This work offers a notable advance in improving the spatial cognition of VLMs, addressing a key limitation for their deployment in geospatial and geographic visual question answering tasks.
Researchers introduce Self-Consistent Flow (SC-Flow), a method that jointly trains a single network to predict both local velocity and data endpoint in rectified-flow generative models. By adding a lightweight consistency loss, SC-Flow unifies the strengths of both parameterizations, stabilizing training and improving the straightness of generation paths. Experiments on image generation tasks show that SC-Flow achieves notable quality improvements over standard rectified-flow baselines with minimal computational overhead.
Why it matters: This work offers a principled approach to combining two key training targets in rectified flow models, potentially enhancing the stability and quality of generative image models.
A new evaluation framework uses four vision-language models (VLMs) to assess EEG-to-image reconstructions, introducing Tolerant Perceptual Alignment Scores (T-PAS) and Tolerant Semantic Alignment Scores (T-SAS). The distilled BCI-Coherence Score (BCS) demonstrates lower mean absolute error and higher correlation with human judgments than traditional pixel or representation-based metrics, addressing the challenge of distinguishing visual fidelity from semantic recoverability.
Why it matters: This framework offers a more meaningful way to evaluate EEG-to-image reconstructions, potentially advancing brain-computer interface research by better capturing both perceptual and semantic accuracy.
Researchers introduce RINO (RGB In and RGB Out), a unified vision model that reformulates diverse visual tasks—including segmentation, depth estimation, and pose-to-image generation—as RGB-to-RGB image editing problems. Using a single architecture and shared parameters, RINO achieves robust zero-shot performance across tasks without task-specific fine-tuning. This approach aims to establish a shared visual interface, analogous to how language models operate over text.
Why it matters: RINO demonstrates a significant step toward unified vision models capable of handling multiple tasks with a single architecture, potentially simplifying deployment and advancing general visual understanding.
Researchers have developed SpikeDS, a spiking neural network architecture that predicts perineural invasion (PNI) in cholangiocarcinoma from 3D MRI scans. SpikeDS employs dual sparsity—combining activation sparsity from binary spike communication and spatial sparsity from window pruning—to reduce computational costs while maintaining diagnostic accuracy. In a clinical cohort study, SpikeDS achieved higher accuracy (AUC 0.753) and greater energy efficiency than existing methods.
Why it matters: This work demonstrates a promising approach for efficient and accurate AI-based medical imaging analysis, which could improve cancer prognosis in clinical practice.
A new preprint investigates how the choice of representation space and reference construction methods affect the performance of training-free synthetic image attribution. The study finds that attribution accuracy is highest when using intermediate layers of CLIP and DINOv2 models, and that semantically constrained references further improve results, particularly when only a small number of references are available. The analysis highlights the importance of both representation selection and reference construction in building effective attribution systems.
Why it matters: The findings offer practical insights for designing scalable, training-free attribution systems that can adapt to new image generators without retraining.
A new preprint compares large language models (LLMs) such as GPT, Twitter-roBERTa, and LLaMA to traditional machine learning methods for analyzing open-ended survey responses. The study finds that LLMs achieve higher classification accuracy, especially in sentiment and thematic analysis, but exhibit significant variation in consistency and the explicitness of their reasoning. These results highlight important trade-offs between predictive performance and interpretability in large-scale qualitative research.
Why it matters: The study offers practical insights for researchers seeking to balance automation with interpretive rigor when applying LLMs to qualitative data analysis.
AgentSociety 2 is a new research environment that integrates large language model (LLM) agents as both AI social scientists and simulated participants, automating the end-to-end workflow of social science research. The system enables hypothesis generation, experiment design, simulation execution, result interpretation, and manuscript drafting across micro, meso, and macro social scenarios. It demonstrates the ability to reproduce qualitative patterns from prior studies and supports large-scale, auditable simulations.
Why it matters: This work advances computational social science by providing a scalable, reproducible, and auditable platform for automating complex social experiments with AI agents.
The GEST-Engine is a system that generates fully-annotated synthetic multi-actor video from natural language input by maintaining an explicit, inspectable world model represented as a Graph of Events in Space and Time (GEST). It produces frame-aligned RGB video, depth, segmentation, pose, and other annotations at zero marginal annotation cost. The system guarantees object permanence and temporal consistency, making it suitable for generating training data and evaluation benchmarks for video understanding.
Why it matters: GEST-Engine enables scalable production of richly annotated synthetic video with guaranteed consistency, potentially reducing reliance on manual annotation in video research.
A new method uses vision-language models (VLMs) to detect anomalous frames between maintenance task videos, enabling automatic extraction of expert-specific actions and contextual decision-making scenes. In simulated maintenance experiments, the approach achieved extraction rates of 65% for actions and 61% for decision-making scenes, outperforming conventional methods. The technique leverages frame-wise visual descriptions and intra-video self-similarity to identify key moments of expert know-how.
Why it matters: This method could facilitate the transfer of expert knowledge to less experienced workers by automatically identifying and extracting critical scenes from maintenance videos.
A new preprint examines the use of large language models (LLMs) for content moderation through 'policy-as-prompt' methods, where moderation policies are given to LLMs as natural-language prompts. The authors argue that this approach introduces specific risks and harms, and that simply writing prompts is not sufficient for effective or meaningful community governance. They propose several considerations for improving prompt governance but conclude that prompt-writing alone cannot ensure robust moderation outcomes.
Why it matters: This research highlights important limitations and governance challenges for AI-driven, prompt-based content moderation systems as their use expands in online communities.
A new preprint compares institutional and course-level generative AI policies at U.S. research-intensive universities. The study finds that while institutions tend to be more supportive of GenAI use, course-level guidance in computing education remains cautious. The authors propose an instructor-centered framework to guide future GenAI adoption in courses.
Why it matters: This research highlights a disconnect between university-wide AI policies and classroom practices, offering a framework to help computing educators navigate GenAI adoption.