AI/ML arXiv cs.AI

Candidate Attended Dialogue State Tracking Using BERT

A novel scalable framework for multi-domain dialogue state tracking using BERT to achieve zero-shot generalization for task-oriented dialogue systems.

AI/ML arXiv cs.AI

Rethinking Quantum Continual Learning with Quantum Fisher Information

Proposes Quantum Elastic Weight Consolidation (QEWC) to mitigate catastrophic forgetting in quantum continual learning using Quantum Fisher Information.

AI/ML arXiv cs.AI

Revisiting data-driven dynamic security assessment with a tabular foundation model

A study demonstrating how tabular foundation models (TFM) can perform dynamic security assessment of power systems via in-context learning with minimal labeled data.

AI/ML arXiv cs.AI

Loop the Loopies!

Introduction of Loopie, a series of looped Transformer MoE models that outperform vanilla Transformers in reasoning and achieve gold-medal performance in IMO and IPhO.

AI/ML arXiv cs.AI

Frontier AI performance across the business disciplines: a case-grounded benchmark of knowledge work and analytical reasoning

Presentation of BusinessCaseBench, a benchmark designed to measure the analytical reasoning and knowledge work capabilities of frontier AI models in business disciplines.

AI/ML arXiv cs.AI

When Model Merging Rivals Joint Multi-Task Reinforcement Learning: A Task-Vector Geometry Analysis

An analysis of model merging in reinforcement learning, finding that merging independently trained specialists can match the performance of jointly trained models.

AI/ML arXiv cs.AI

Spatial Normalization for Cross-Domain Retinal Layer Segmentation in Optical Coherence Tomography

Research on the use of spatial normalization as a preprocessing strategy to improve the robustness of retinal layer segmentation in Optical Coherence Tomography.

Cybersecurity Hacker News

Exploit brokers pay $500k for WordPress RCEs. I found one with GPT5.6 and $25

A user claims to have discovered a WordPress Remote Code Execution (RCE) vulnerability using GPT-5 and a small budget, highlighting the potential for AI to lower the barrier for finding high-value exploits.

Software Engineering Hacker News

Eliminating Go bounds checks with unsafe

A technical discussion on how to eliminate Go bounds checks using the 'unsafe' package to improve performance.

Tech Business/VC Hacker News

EU Exempts Apple Watch and AirPods from Battery Removal Requirements

The EU has exempted Apple Watch and AirPods from new battery removal requirements, providing a reprieve for Apple regarding small wearable device designs.

AI/ML arXiv cs.AI

DECODEM: Data Extraction from Corporate Organizational Documents via Enhanced Methods

The DECODEM paper introduces benchmark datasets and evaluates LLM pipelines for automating the extraction of structured governance variables from unstructured corporate documents.

AI/ML arXiv cs.AI

Perceived AGI: Believability as Dimensional Completeness, Not Capability

This conceptual framework proposes that AI believability stems from 'dimensional completeness' (time, truth, entropy, love) rather than task capability.

AI/ML arXiv cs.AI

Induction in Both Directions: A Mechanistic Analysis of In-Context Learning in Masked Diffusion Language Models

A mechanistic analysis of Diffusion Language Models (DLMs) reveals they implement bidirectional induction circuits for in-context learning, unlike the unidirectional approach of autoregressive transformers.

AI/ML arXiv cs.AI

Orbis 2: A Hierarchical World Model for Driving

Orbis 2 is a hierarchical world model for driving that uses a two-stage training paradigm (diffusion forcing then teacher forcing) to balance representation quality and rollout stability.

AI/ML arXiv cs.AI

On the Failure of Boundary-Seeking Distillation in Bottlenecked Generative Architectures

Researchers demonstrate that boundary-seeking distillation fails in bottlenecked generative architectures like autoencoders because it violates the latent manifold geometry.

AI/ML arXiv cs.AI

When Not to Automate: A Formal Protocol for Human Preservation in AI-Optimized Organizations

The PHP-AIO protocol provides a formal framework to quantify systemic risks—such as tacit knowledge loss—to determine when AI automation should be avoided in organizations.

Other arXiv cs.AI

Sociocultural Influences on Opinion Formation: Word of Mouth Dynamics, Mass Media and Behavioural Development

A study on how word-of-mouth dynamics and mass media influence opinion formation and social attitude development within diverse sociocultural groups.

AI/ML arXiv cs.AI

Modularized Dynamic-Granularity Video LLM for Multi-Event Long Video Understanding

Researchers introduce MoD-VLLM, a framework that uses modularized dynamic-granularity encoding to better understand multiple events in long videos.

AI/ML arXiv cs.AI

On the Geometry of Learned Representations in Event-Based Multi-Modal Egomotion Estimation

This study analyzes the latent space geometry of multi-modal networks used for event-based egomotion estimation, bridging data-driven fusion with analytical theory.

AI/ML arXiv cs.AI

Knowledge-Assisted Multi-Graph Dependency Learning for Multivariate Time Series Anomaly Detection in Multi-Stage Industrial Processes

A new multi-graph framework for multivariate time series anomaly detection in industrial processes incorporates process knowledge to improve sensor dependency modeling.