Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2608.22690v1 Announce Type: new Abstract: Cell Painting microscopy captures how cells change after a chemical or genetic perturbation. Connecting these images to the perturbations that produced...
arXiv:2608.23100v1 Announce Type: cross Abstract: Robot co-design via bi-level optimization couples within-lifetime controller learning for fitness evaluation with cross-generational morphological ev...
The paper investigates how large language models (LLMs) hallucinate nonexistent software packages during code generation and evaluates methods to mitigate this issue. It finds that current evaluation practices overestimate hallucination rates, especially for Python, and that Retrieval-Augmented Generation (RAG) and Self-Refine reduce hallucinations across multiple models and languages. The study also introduces Package Utility (PU) to measure whether defenses preserve useful recommendations and shows that Greedy decoding offers the best trade‑off between mitigation and utility, while adversarial prompts significantly increase hallucination rates, particularly in Ruby.
The paper introduces Align‑LoRA, a unified LoRA framework for multi‑task learning that replaces complex, isolated adapter designs with a single‑adapter model enhanced by a higher rank and an explicit alignment loss. It demonstrates that a router‑free, multi‑head model with high inter‑head redundancy can outperform more elaborate baselines, and that a unified LoRA can achieve competitive performance while enabling weight merging and zero inference latency. Extensive experiments and theoretical analysis confirm that Align‑LoRA surpasses prevailing approaches, offering a simpler, production‑friendly paradigm for parameter‑efficient fine‑tuning of large language models.
arXiv:2608.21606v1 Announce Type: new Abstract: Machine unlearning aims to remove the influence of targeted training data from a model while preserving its remaining capabilities, but evaluating whet...
arXiv:2608.16466v2 Announce Type: replace-cross Abstract: Bayesian optimal experimental design (BOED) aims to collect informative data by optimizing an expected utility reflecting the goals of an exp...
arXiv:2607.02486v2 Announce Type: replace Abstract: Descriptor-free visual localization eliminates high-dimensional descriptor storage, preserves scene privacy, and simplifies map maintenance, yet it...
arXiv:2608.21836v1 Announce Type: new Abstract: Large language models have become increasingly capable agents for low-level code and kernel optimization, but isolated kernel benchmarks provide only a...
arXiv:2608.10503v2 Announce Type: replace Abstract: As Large Language Models (LLMs) are increasingly deployed as autonomous agents, accurately evaluating their latent values and biases is critical. T...
arXiv:2608.22329v1 Announce Type: cross Abstract: Emotion-aware artistic image generation requires a model to satisfy semantic content, artistic style, and target emotion simultaneously. The key chal...
The paper studies how midtraining with synthetic documents can align AI models to the value of animal compassion. It introduces ANIMA, a 26‑question benchmark covering 13 ethical dimensions, and shows that training on 3,000 documents yields 77% accuracy versus 40% for instruction‑tuning. The improvement disappears after 5,000 samples, indicating that document‑based interventions may need explicit preservation strategies.
arXiv:2608.21914v1 Announce Type: new Abstract: We align three upper ontologies that sit under UK and NATO defence data infrastructure: the Information Exchange Standard (IES), the Higher Quality Dat...
The paper introduces a framework to tackle modality imbalance in multimodal learning by focusing on sample-level variations. It defines a Modality Gap metric to measure prediction discrepancies, models the resulting bimodal distribution with a Gaussian Mixture Model, and uses Bayesian probabilities for soft separation of balanced and imbalanced samples. A two‑stage training process—Warm‑up and Adaptive Training—reallocates loss weights based on the GMM, strengthening alignment for imbalanced samples while favoring fusion for balanced ones, and shows superior performance over existing baselines.
arXiv:2605.11533v4 Announce Type: replace Abstract: Routine clinical check-up reports combine laboratory measurements, physiological assessments, imaging findings and visually structured information,...
The paper introduces RASET, a router‑agnostic safety‑critical expert tuning framework for Mixture‑of‑Experts (MoE) large language models. RASET identifies a small subset of experts that are responsible for safety enforcement and applies parameter‑efficient tuning only to those experts, preserving the model’s intrinsic routing behavior. Experiments on five open‑weight MoE backbones show that RASET achieves a high safety‑bypass yield, outperforming existing baselines by a significant margin.
arXiv:2608.22068v1 Announce Type: new Abstract: Geothermal well arrays, which organize multiple geothermal wells into carefully planned geometric configurations, provide opportunities to enhance ener...
arXiv:2608.22960v1 Announce Type: new Abstract: Coding agents are increasingly evaluated not only by whether they solve a task, but also by how they execute it. However, existing process-level evalua...
arXiv:2608.22024v1 Announce Type: cross Abstract: Estimating heterogeneous single and interaction treatment effects from observational data under multiple simultaneous treatments is crucial for decis...
GRACE is a step‑level benchmark for evaluating the faithfulness of chain‑of‑thought reasoning over context. It provides human annotations for each step in CoT traces from 10 models across 4 datasets, labeling faithfulness, error category, and natural‑language explanations. The benchmark introduces a data‑driven taxonomy that splits errors into GRACE‑Inference (deductive) and GRACE‑Grounding (factual) tracks, each with four categories, and demonstrates that incorporating step‑level faithfulness signals can improve downstream accuracy and reasoning reliability.
The paper introduces a new type of adversarial attack on automated fact‑checking systems that uses large language models to rephrase claims with persuasive techniques. By applying 15 persuasion methods across five categories, the authors evaluate how these rewrites affect claim verification and evidence retrieval on the FEVER and FEVEROUS benchmarks. Results show that persuasive rewrites significantly degrade both verification accuracy and evidence retrieval performance, underscoring the vulnerability of current fact‑checking systems to such attacks.