Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2610.07936v1 Announce Type: new Abstract: Systematicity, the probabilistic mapping of form to meaning, permeates language at all levels, and sublexical cues have been shown to govern human pseu...
arXiv:2610.08026v1 Announce Type: new Abstract: In recent years, several methods for detecting when large language models (LLMs) hallucinate have been developed. These methods are often benchmarked w...
arXiv:2610.08303v1 Announce Type: new Abstract: Current evaluation of multilingual Large Language Models (LLMs) rests on an implicit Translation-Isomorphism Assumption (TIA): that semantic structures...
arXiv:2507.02115v3 Announce Type: cross Abstract: Synthesizing second-language (L2) speech is potentially highly valued for L2 language learning experience and feedback. However, due to the lack of L...
The study introduces MedQADE, a German open‑response clinical benchmark with 3,800 question‑answer pairs and physician reference annotations. It evaluates large language models (LLMs) as judges, finding that while some LLMs (e.g., Gemini 3 Flash) achieve physician‑level agreement on correctness, they exhibit self‑bias and low abstention rates. Physicians showed moderate agreement on correctness but limited agreement on difficulty, and their abstention increased with perceived difficulty.
The paper investigates how diffusion models, even after fine‑tuning to unlearn harmful concepts, still retain a linear residual of those concepts in token embeddings. It introduces SubAttack, a jailbreak method that extracts this residual by learning orthogonal attack tokens composed of interpretable textual elements, and SubDefense, a plug‑and‑play defense that projects out the same subspace to suppress the residual. Experiments show SubAttack outperforms prior attacks in power and transferability, while SubDefense offers stronger robustness and better preservation of safe generation quality.
The paper introduces Latent-Action-Guided Video-Language Feature Learning (LAG-VLFL) for recognizing surgical instrument–tissue interactions. By compressing frame-to-frame feature changes into latent actions and predicting next‑frame features, the method aligns video and textual action descriptions without requiring extra spatial or motion annotations. Experiments show that LAG-VLFL improves interaction grounding, temporal‑direction sensitivity, and achieves competitive recognition with faster inference and lower INT4 accuracy loss compared to V‑JEPA2/2.1.
arXiv:2610.06896v1 Announce Type: new Abstract: Frontier multimodal large language models (MLLMs) are increasingly positioned as general purpose visual reasoners as part of the quest for artificial g...
arXiv:2610.06945v1 Announce Type: new Abstract: Mechanistic interpretability often relies on the Linear Representation Hypothesis (LRH), which assumes that high-level concepts are encoded as linear d...
arXiv:2610.07689v1 Announce Type: new Abstract: Vision-Language Models (VLMs) such as CLIP excel in global semantic alignment but often lack fine-grained perceptual capabilities. This hinders dense p...
arXiv:2610.07793v1 Announce Type: new Abstract: Non-contact three-dimensional reconstruction of thin, sheet-like heritage artifacts poses significant geometric and registration challenges. Due to the...
arXiv:2610.07903v1 Announce Type: new Abstract: Adapting vision-language models to downstream tasks has achieved remarkable success by leveraging pseudo-labels generated from unlabeled data. Existing...
arXiv:2610.08419v1 Announce Type: new Abstract: Slice-to-volume registration between ultrasound (US) and preoperative computed tomography (CT) imaging would enhance many minimally invasive interventi...
arXiv:2610.08539v1 Announce Type: new Abstract: Remote sensing scene classification is a fundamental task in Earth observation and geospatial analysis. Existing approaches mainly follow three paradig...
arXiv:2610.08573v1 Announce Type: new Abstract: Cooperative perception improves autonomous driving by sharing complementary observations among vehicles and roadside infrastructure for 3D object detec...
arXiv:2604.09045v2 Announce Type: replace Abstract: Recent works on 3D scene understanding leverage 2D masks from visual foundation models (VFMs) to supervise radiance fields, enabling instance-level...
The paper introduces Constant‑Curvature Sliced Gromov‑Wasserstein (CCSGW), a new divergence for aligning probability distributions on heterogeneous constant‑curvature spaces such as hyperbolic and spherical manifolds. It extends sliced Gromov‑Wasserstein by adding geodesic‑based one‑dimensional projections for spherical spaces, enabling efficient and principled comparison across manifolds with different curvatures while preserving intrinsic geometric relationships. The authors provide theoretical analysis showing that CCSGW controls intrinsic geometric discrepancy and demonstrate consistent performance gains when integrated into mixed‑curvature learning tasks like graph anomaly detection, node classification, and multimodal learning.
The paper introduces “GraphPDHG,” a message‑passing neural network that aligns with the Chambolle‑Pock Primal–Dual Hybrid Gradient algorithm to solve general graph saddle‑point problems. The authors prove that “GraphPDHG” can efficiently simulate PDHG and can learn an accelerated version of the algorithm. Experiments demonstrate that the learned warm start improves second‑order optimizers and that alignment with PDHG yields better size generalization compared to standard GNN baselines.
The study examines why improved traffic forecasts do not always lead to better signal‑control outcomes. Using 29 days of reconstructed traffic data from Xuancheng, China, the authors evaluate various forecasting methods—including point forecasts, conformal intervals, and dependence‑aware scenarios—within a closed‑loop control framework. While entry‑level and movement‑level forecasts reduce mean absolute error by about 4%, the resulting signal‑control decisions sometimes worsen queue vehicle‑seconds and spillback exposure, highlighting gaps between forecast accuracy and operational benefit.
The paper investigates whether semantic entropy—a measure of disagreement among a language model’s sampled answers—can serve as a cheap signal for deciding when to route a query from a small to a larger language model. Experiments on GSM8K and other benchmarks show that semantic entropy can distinguish small‑model mistakes and improve routed accuracy, but the authors also reveal that a simple question‑difficulty rule can mimic its performance and that other factors (definition of success, benchmark design, live sampling cost) can undermine its effectiveness. They propose a checklist of checks to validate escalation signals and demonstrate how to predict when a cached‑outcome approach will fail. "whyItMatters":"The study highlights the importance of rigorous evaluation of escalation signals, showing that seemingly promising metrics can be misleading without proper controls and that practical routing decisions must account for cost and benchmark design."