AI/ML arXiv cs.AI

Proxy Exploration and Reusable Guidance: A Modular LLM Post-Training Paradigm via Proxy-Guided Update Signals

PUST is a post-training framework that decouples update-signal exploration from distribution alignment using proxy models.

AI/ML arXiv cs.AI

CDFM: Towards a General-Purpose Causal Discovery Foundation Model

CDFM presents a general-purpose foundation model for zero-shot structural causal discovery from observational data.

AI/ML arXiv cs.AI

Toward Inclusive Avatar Design with Limb Differences Through Artificial Intelligence

A review of AI-driven technical approaches for creating inclusive 3D avatars that represent people with limb differences.

Other Hacker News

Global Warming at 3 °C by 2050? What's Behind the New German Climate Warning

A discussion regarding a German climate warning predicting global warming could reach 3°C by 2050.

Software Engineering Hacker News

Ambient Website Background Clouds

A post about implementing ambient background clouds for websites.

AI/ML arXiv cs.AI

Longitudinal Multi-View Breast Cancer Risk Prediction

Researchers propose LMV-Net, a deep learning model for longitudinal multi-view breast cancer risk prediction using mammography views.

Cybersecurity arXiv cs.AI

Understanding the Impact of AI Code Assistants on Security API Usage: An Empirical Study

An empirical study finds that AI code assistants like GitHub Copilot improve functional correctness but do not significantly improve the secure usage of security APIs.

AI/ML arXiv cs.AI

Characterising AI Models for Cataloguing

An evaluation of AI models for the task of creating digital catalogue records for digital collections.

AI/ML arXiv cs.AI

Beyond Sally-Anne: Evaluating Theory of Mind in LLMs using Epistemic Schelling Points

Introduction of the Epistemic Asymmetry Schelling Task (EAST) to evaluate Theory of Mind in LLMs through dialogue games.

AI/ML arXiv cs.AI

BackgroundMellow: A Multi-Modal Cohesive Framework for Narrative-Driven Rich Cinematic Soundscape Generation

BackgroundMellow is a multi-modal framework for generating synchronized, narrative-driven cinematic soundscapes using a master-specialist agent architecture.

AI/ML arXiv cs.AI

A Glimpse into Long-term Physical Coexistence with Intelligent Robots

Introduction of PHILIA, a multi-robot agent architecture that decouples high-level reasoning from low-level execution for long-term physical coexistence.

AI/ML arXiv cs.AI

Agentic Routing: The Harness-Native Data Flywheel

Proposes Harness-Native agentic routing to optimize model selection within AI agent execution harnesses to create a data flywheel for training.

AI/ML arXiv cs.AI

Uncertainty Quantification for EO Regression Tasks: Building Height, Tree Canopy Height and Above-ground Biomass Estimation

A study on uncertainty quantification for Earth Observation regression tasks including building height and biomass estimation using Sentinel data.

AI/ML arXiv cs.AI

Pix2Act: Image-Space Manipulation Policies with Equivariant Augmentation

Pix2Act is an imitation learning method for 3D manipulation that simplifies control by predicting 2D image-space trajectories and recovering poses via triangulation.

AI/ML arXiv cs.AI

Enhancing LLMs through human feedback: a journey towards self-improvement

The study proposes enhancing RAG systems by integrating an auxiliary feedback RAG system and a human-in-the-loop workflow for continuous self-improvement.

AI/ML arXiv cs.AI

Towards Predictive, Aligned, and Scalable Robot Learning

Lumo-2 is a latent world-action model that uses a multi-stage modality pre-alignment strategy to improve robotic predictive reasoning and control performance.

AI/ML arXiv cs.AI

Automated Textbook Auditing with Multi-Agent LLM Systems

The AI Textbook Auditor is a multi-agent LLM pipeline designed to automate factual and linguistic auditing of educational materials.

AI/ML arXiv cs.AI

A Unified Framework for Comprehensive Cardiac CT Segmentation and Phenotyping: Human-in-the-Loop Data Annotation, Vision Foundation Model Development, Multicenter Evaluation and Clinical Validation

A unified framework for cardiac CT segmentation using a self-supervised foundation model and a large expert-annotated dataset to improve clinical phenotyping.

Cybersecurity arXiv cs.AI

Mako: A Self-Evolving Agentic Operating System (SE-AOS) for Autonomous Web Exploitation

Mako is a self-evolving agentic OS for autonomous web exploitation that treats exploit capabilities as a mutable kernel it can extend at runtime.

AI/ML arXiv cs.AI

The Paternalistic Filter: Epistemic Injustice and Differential Refusal in LLM-Mediated History Education for Marginalized Romanian Students

An audit of LLMs acting as history tutors reveals systemic biases and 'epistemic paternalism' that disadvantage marginalized students.