Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanis...
Prior work has shown that neural networks exhibit implicit biases toward low-complexity structure (e.g., spectral bias), memorization dynamics, and compression-like effects during training, but a unif...
The article reports that on a set of 100 randomly selected tasks from an internal Binary Exploitation benchmark, GLM‑5.3 achieved full control‑flow hijacks in 4% of the trials, while Claude Mythos Preview did so in 6%. Both models outperform earlier versions such as Claude Opus 4.6 and GLM‑5.2, which succeeded in none of the trials. This indicates that a significant threshold in adversarial exploitation capabilities has been crossed by the newer models.
Training a model to generate visual content can encourage it to learn rich perceptual capabilities related to geometry, spatial relationships, and objectness; yet, its benefits for visual understandin...
Generative models for function-valued data, such as time series and solutions of partial differential equations, must learn distributions over infinite-dimensional spaces. Functional Flow Matching (FF...
Embodied systems must make knowledge acquired during one encounter usable in another despite changes in viewpoint, motion, and illumination. Yet aggregate cross-video accuracy conflates failures of lo...
Concept Bottleneck Models (CBMs) built on vision-language models such as CLIP represent a latent space as human-understandable concepts. These representations are unfaithful: related concepts are enta...
Rendering accurate Chinese text remains challenging for text-to-image models. Existing OCR-based reinforcement-learning rewards compare decoded transcripts with target strings. Such rewards overlook t...
Audio-visual large language models (AVLLMs) have made remarkable progress in multimodal understanding and reasoning through interactions among visual, auditory, and linguistic information. However, re...
The article "How to Design Architectural Guardrails Around AI Agents" discusses essential agent design patterns that data engineers should understand. It emphasizes the importance of establishing clear architectural boundaries to ensure AI agents operate safely and effectively within larger systems. The post was originally published on Towards Data Science.
Balancing performance trade-offs on long-tailed data distributions remains a long-standing challenge in visual recognition. Existing methods mainly improve tail classes through re-balancing, represent...
While Large Vision-Language Models (LVLMs) achieve remarkable success, hallucinations remain a significant barrier to their reliable deployment. Recent studies primarily attribute these issues to cros...
Structured images, such as diagrams, charts, and flowcharts, are inherently symbolic and can be compactly represented in an editable format, yet in practice, they are often rendered as images, and the...
Fine-tuning-as-a-service lets users adapt a safety-aligned language model to their own data, but it also creates a harmful fine-tuning attack surface: a small amount of harmful data mixed into an othe...
One-dimensional (1D) variable-length visual tokenizers enable adaptive compression by varying the number of tokens, allowing downstream autoregressive (AR) models to flexibly trade off generation qual...
Multi-agent LLM systems are increasingly structured like organizations, with roles, critics and shared memory, yet they are evaluated on tasks that last minutes. We ask how such an organization behave...
The paper presents MoCoP v2, an enhanced contrastive pretraining method that aligns small molecule embeddings with deep‑learning‑derived cell morphology profiles. By replacing CellProfiler fingerprints with richer image‑encoded features, the new embeddings better capture how molecules alter cell morphology, leading to improved QSAR, toxicity, ADME, and activity predictions. Performance scales log‑linearly with training data size, indicating further gains with larger datasets.
GyroNovo is a new framework for de novo peptide sequencing that improves fragment imputation by guiding the process with decoder errors observed during training. It introduces mass-aware attention using rotary embeddings to encode pairwise mass differences between spectral peaks, and creates easy and hard augmented views of spectra to train the decoder under varying corruption levels. Experiments on NovoBench demonstrate significant gains, with about 9 percentage points higher peptide-level precision and 7 percentage points higher amino-acid-level precision compared to the state-of-the-art baseline.
arXiv:2609. 30634v1 Announce Type: new Abstract: How many assignments can a language model recall before it loses track of which value belongs to which entity?
The paper addresses the failure of online conformal prediction when predictions influence actions that determine which outcomes are used for calibration. It introduces Propensity-Weighted Online Conformal Prediction (PW‑OCP), an inverse‑propensity‑weighted recursion that debiases calibration, and a doubly robust variant (DR‑OCP) that further reduces bias. Experiments on synthetic decision tasks, open bandit data, and financial rebalancing demonstrate that PW‑OCP and DR‑OCP improve counterfactual coverage and downstream regret while preserving prediction‑set sharpness.