Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2609.25505v1 Announce Type: cross Abstract: Rapid intensification (RI) remains one of the most consequential and difficult aspects of tropical cyclone (TC) forecasting. Although full-physics nu...
arXiv:2609.26069v1 Announce Type: cross Abstract: Simulation-based tutor selection can be unstable when predictively adequate learner models imply different policy rankings. RankCert certifies one of...
arXiv:2504.20106v4 Announce Type: replace Abstract: Ensuring that large language models (LLMs) are both helpful and harmless is a critical challenge, as overly strict constraints can lead to excessiv...
arXiv:2508.20443v3 Announce Type: replace Abstract: Large language models (LLMs) are trained on massive datasets that may include private or copyrighted content. Due to growing privacy and ownership...
arXiv:2608.01069v2 Announce Type: replace Abstract: Bandit algorithms generate data for downstream inference, but adaptive sampling biases post-bandit sample means. We analyze this bias for stable in...
arXiv:2605.01047v2 Announce Type: replace-cross Abstract: Hallucinations remain an unsolved problem for LLMs, and package hallucinations are a particularly dangerous instance of this phenomenon. Pack...
arXiv:2609.25056v1 Announce Type: new Abstract: A feedback-based word deduction framework based on the Jotto problem is proposed, and the problem space is represented as a weighted graph where all va...
arXiv:2609.25066v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to simulate human survey responses and behavioral reactions, yet unreliable simulations can mislead...
arXiv:2609.26399v1 Announce Type: new Abstract: Scene safety understanding plays a life-or-death role in situational awareness in various critical domains. Traditional methods that rely on learning d...
arXiv:2609.26185v1 Announce Type: cross Abstract: Large Language Models (LLMs) demonstrate impressive capabilities across many applications but remain vulnerable to jailbreak attacks, which elicit ha...
arXiv:2609.26056v1 Announce Type: new Abstract: Vision-Language Models (VLMs) offer promising capabilities for automated sports coaching but face a fundamental limitation: they implicitly compare aga...
arXiv:2609.26274v1 Announce Type: new Abstract: The rapid evolution of generative AI (e.g., Sora, Hunyuan) makes it essential to develop effective detection strategies that can generalize across ever...
arXiv:2609.26623v1 Announce Type: new Abstract: Diffusion-based synthetic data generation offers a promising route for sharing medical imaging data without releasing sensitive patient records. Howeve...
arXiv:2609.23796v2 Announce Type: replace Abstract: Single-image 3D object generation can now produce high-fidelity assets, yet accurately placing them into a coherent scene layout remains an open ch...
The paper presents a decision‑aware framework for predicting dementia‑related crash severity that emphasizes auditability and selective deferral. Using 4,781 Texas crash records, the authors evaluate several models—including structured, narrative, fusion, calibrated fusion, BERT‑family, and local large‑language‑model baselines—under a stratified 70/15/15 split. The leakage‑controlled Gemma model achieves the highest macro‑F1 of 0.545, while a calibrated fusion model reaches 0.522 macro‑F1 with an expected calibration error of 0.033; selective deferral further improves performance, raising macro‑F1 to 0.573 at 70% coverage and reducing severity cost to 0.577.
The paper discusses the need to adapt incident reporting frameworks for AI agents, which are rapidly deployed and face unique security challenges. By comparing AI systems and agents and consulting 23 experts, the authors identify key reporting elements such as agent memory, autonomy levels, and tool usage. They also highlight open research questions, potential reporting weaknesses like data leakage, and outline privacy requirements for secure AI agent deployment.
The paper offers a unified probabilistic framework for large language models, describing them as probability measures over token sequences defined by autoregressive conditional distributions. Training is cast as maximum‑likelihood estimation solved via stochastic gradient methods, while generation is treated as sequential simulation of the resulting stochastic process. It also explores how the asymmetry of the Kullback–Leibler divergence relates to hallucination and the distinction between plausibility and truth, and extends the perspective to diffusion models that generate data by simulating a reverse‑time stochastic process.
FairMean is a new approach for distributed learning that addresses the conflict between fairness and robustness to label poisoning attacks. It assigns weights to client gradients based on a bounded, nondecreasing function of local loss, giving higher weight to high‑loss clients to promote fairness while limiting the influence of poisoned clients. The method is shown to improve fairness compared to standard average‑loss minimization and to reduce accuracy variance while boosting worst‑client accuracy in experiments.
The paper presents an empirical audit of privacy leakage in edge‑private graph generators by testing whether outputs from edge‑neighbouring inputs remain distinguishable. It introduces statistically valid lower bounds on privacy loss and compares direct‑edge, local‑structural, and GNN‑based attacks based on the geometry around a target edge. Experiments on two generators and two networks reveal that privacy leakage varies with both the mechanism and the network, and that learned representations expose information beyond conventional local statistics.
KwaiMind is a commercial image editing system that combines general editing capabilities with e-commerce specialization. It uses an agent-based data engine with 1.8 million editing pairs and a multimodal diffusion transformer trained through pre‑training, fine‑tuning, preference optimization, and online reinforcement learning. The system is guided by a vision‑language judge and specialized rewards for click‑through rate, text rendering, and product consistency, and it achieves top scores on ImgEdit, GEdit, REDEdit, and the new Ecom‑Bench, while improving predicted and actual CTR in offline and online experiments.