Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2610.32107v1 Announce Type: new Abstract: Participating in a conversation involves not only deciding what to say but also whether to speak at all. This judgment carries even greater significanc...
arXiv:2610.32309v1 Announce Type: new Abstract: Why can a task-preserving rewrite turn a correct answer into a failure while a larger internal change leaves the answer intact? We develop a direction-...
arXiv:2610.32415v1 Announce Type: new Abstract: LLM-based annotation has advanced rapidly, yet existing benchmarks remain misaligned with real-world practice. Actual annotation often involves lengthy...
arXiv:2610.32835v1 Announce Type: cross Abstract: Modern vision encoders are widely used as semantic representations, but the relationship between embedding similarity and human-visible semantics can...
arXiv:2608.09551v2 Announce Type: replace Abstract: Human language is grounded in pragmatics, necessitating implicit (extralinguistic) context to interpret language, e.g., world knowledge and social...
arXiv:2608.23818v2 Announce Type: replace Abstract: Transformer-based language models are widely used as models of human language processing, yet their attention mechanisms allow lossless access to t...
arXiv:2610.31898v1 Announce Type: new Abstract: Traffic safety videos provide rich evidence for understanding interactions between pedestrians and vehicles, but synthetic-to-real evaluation makes thi...
arXiv:2610.32230v1 Announce Type: new Abstract: Modern autonomous driving systems widely adopt multi-view Bird's-Eye-View (BEV) based 3D perception models due to their superior performance. Despite t...
arXiv:2610.32317v1 Announce Type: new Abstract: Video moment retrieval aims to localize a query-relevant temporal segment in an untrimmed video. Existing methods typically perform video-text alignmen...
arXiv:2503.10212v3 Announce Type: replace Abstract: Analyzing animal behavior is crucial in neuroscience, yet deciphering its intricate dynamics remains challenging. Traditional machine vision approa...
arXiv:2412.08484v5 Announce Type: replace-cross Abstract: Modern mesh generation pipelines whether learning-based or classical often produce outputs requiring post-processing to achieve production-qu...
arXiv:2506.19186v2 Announce Type: replace-cross Abstract: We study the importance-tempered random-walk Metropolis (RWM) algorithm on $\mathbb{R}^d$, which targets the tempered density $\pi_\beta(x) \...
ProfileSim is a profile‑driven user simulation framework designed for financial marketing outbound calls. It jointly models user demographics, Big Five personality traits, and business intents reverse‑engineered from real call logs, allowing fine‑grained control over simulated behavior. The system is fine‑tuned on real voice dialogue data and further refined with Group Relative Policy Optimization using hierarchical rewards, achieving high colloquial naturalness, strong user profile adherence, and low agent scores on long‑tail intents while producing shorter user utterances.
MCPHunt is a benchmark framework that isolates non‑adversarial credential propagation across multi‑server MCP trust boundaries. It uses canary‑based taint tracking, a controlled coverage design with risky, benign, and hard‑negative conditions, and completion‑requires‑secret stratification to separate faithful from policy‑violating propagation. Across 3,615 traces from 5 models, policy‑violating propagation rates range from 11.5 % to 41.3 %, with mitigation via prompting reducing rates to 0.3 % while retaining 80.5 % utility.
The paper introduces Poly‑Ripple, a collaborative in‑context learning framework for indoor radio map estimation that leverages large autoregressive vision models without updating them. Clients provide private demonstrations to predict radio maps for shared scene queries, using ripple‑ordered autoregressive modeling to reflect wireless propagation from the transmitter. Importance‑aware aggregation combines heterogeneous client knowledge within context length limits, and experiments on a public benchmark show Poly‑Ripple outperforms existing ICL baselines.
arXiv:2610.31889v1 Announce Type: cross Abstract: Energy-Based Models (EBMs) are a flexible class of generative models that decouple probabilistic modeling from the generation process. This decouplin...
The paper introduces S‑SALAAD, a framework that jointly learns model weights, low‑rank, and structured sparse components during training. It supports hardware‑compatible sparsity patterns and couples structure learning with budget‑aware compression and dedicated CPU/GPU kernels, allowing efficient inference from a single checkpoint across multiple memory budgets. Experiments on Llama2 and Qwen3 demonstrate improved perplexity, memory usage, and latency on various hardware platforms compared to baselines.
The paper presents a mathematical theory of the Jacobian lens (J‑lens), an interpretability tool that identifies linear directions in language model hidden states resembling working‑memory representations. It shows that the J‑lens preferentially detects coherent features—directions that act as leading, spatially coherent Lyapunov modes—whose perturbations amplify across layers, remain stable across token positions, and are consistent across contexts. The authors validate their theory on a Bayes filter and a toy transformer trained on a Hidden Markov Model, demonstrating that task‑parameter‑induced changes in feature coherence predictably alter J‑lens responses, and introduce a family of ω‑J‑lenses that emphasize different temporal scales of information processing.
The paper introduces TopKLoRA, an interpretable parameter‑efficient fine‑tuning method that activates only the top‑k adapter latents per token. Using this sparse LoRA, the authors fine‑tune large language models on a poisoned dataset, identify and ablate the minimal circuit responsible for a backdoor, and show that the resulting circuits are an order of magnitude smaller than dense LoRA baselines. They also demonstrate how to construct model organisms with localized circuits via gradient‑routing, supporting the development of interpretable‑by‑design architectures.
arXiv:2610.31695v1 Announce Type: cross Abstract: Safety guards check content before an LLM application acts on it, and most paste that content into a plain-text prompt. Content that imitates the pro...