AI/ML arXiv cs.AI

Empathy as Predictive Misalignment Tolerance: A Co-Regulation Framework and the Regime Structure of Dialogue Repair

A theoretical framework reframing AI empathy as 'predictive misalignment tolerance' to better manage divergence in extended dialogues.

AI/ML arXiv cs.AI

How Does Empowering Users with Greater System Control Affect News Filter Bubbles?

A study on how giving users more control over recommendation system parameters can help them recognize and mitigate news filter bubbles.

AI/ML arXiv cs.AI

Structure of the Circular-Dyadic Convolution Error

A mathematical analysis of the algebraic error introduced when substituting the discrete Fourier transform with the Hadamard transform in convolutions.

AI/ML arXiv cs.AI

AV-JEPA: Extending LeJEPA to Audio-Visual Self-Supervised Learning

AV-JEPA introduces a multimodal extension of LeJEPA for audio-visual self-supervised learning without the need for complex decoders or contrastive losses.

AI/ML arXiv cs.AI

Data-driven Video Codec with Implicit Neural Representations

A novel video codec that stores audio-visual data as weights of a sinusoidal representation network (SIREN) and compresses them using quantization and LZMA2.

Hardware/Chips arXiv cs.AI

Lazy Arithmetic using Systolic Arrays for Closing the Verification Gap on Embedded Systems

A proposed hardware and software approach using systolic arrays and left-to-right arithmetic to ensure sound, real-time adaptive-precision quantization for edge AI.

AI/ML arXiv cs.AI

Large Language Models as Unified Multimodal Learners for Clinical Prediction

Demonstrates that converting multimodal clinical data into a single natural language sequence for LLM fine-tuning outperforms specialized fusion architectures.

Other Hacker News

Korg PS-3300 - one of the rarest synthesizers in music history is back

Korg is re-releasing a rare vintage synthesizer, the PS-3300, bringing back a high-end piece of music history.

Tech Business/VC Hacker News

Big Tech Needs to Justify AI Spending as Investors Dump Stocks

Investors are pressuring Big Tech companies to justify the massive capital expenditures on AI infrastructure as stock prices fluctuate.

AI/ML arXiv cs.AI

Behavioral Controllability of Agentic Models for Information Extraction: From Fixed Workflows to Reflective Agents

Researchers study the effectiveness of reflective agents versus fixed LLM workflows for structured information extraction from scholarly PDFs.

AI/ML arXiv cs.AI

NeurOWL: An LLM-Based Neural-symbolic Framework for Incomplete OWL Ontology Reasoning

NeurOWL is introduced as a neuro-symbolic framework to handle reasoning in incomplete OWL ontologies using LLMs and ontology embeddings.

AI/ML arXiv cs.AI

AgentFAIR: A Multi-Agent Collaborative Framework for FAIRness Evaluation of Geospatial Datasets

AgentFAIR is a multi-agent framework designed to evaluate the FAIRness (Findability, Accessibility, Interoperability, Reusability) of geospatial datasets.

AI/ML arXiv cs.AI

Knowledge-Centric Agents for Workflow Generation

A new knowledge-centric framework for generating visual creation workflows (e.g., for ComfyUI) using knowledge inversion and injection.

AI/ML arXiv cs.AI

DSWorld: A Data Science World Model for Efficient Autonomous Agents

DSWorld introduces a world model for data science to predict environment state transitions, significantly accelerating agent training and inference.

AI/ML arXiv cs.AI

A Formally Grounded ODRL Evaluator: Implementation and Comparison

Researchers implement a formally grounded ODRL evaluator to provide consistent and interoperable policy evaluation for data access in European dataspaces.

AI/ML arXiv cs.AI

Closing the AI Trust Gap: The Case for Independent Certification for Trustworthy AI

A proposal for independent certification of trustworthy AI to close the gap between internal corporate responsible AI practices and external verifiable outcomes.

AI/ML arXiv cs.AI

SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery

SciForge is an open-source, multimodal AI workbench designed for scientific discovery, integrating agentic research sprints and evidence governance.

Tech Business/VC Hacker News

Power companies are using eminent domain to seize land for data centers

Power companies are using eminent domain to acquire land for the construction of data centers.

Tech Business/VC Hacker News

Big Tech Is Now Targeting Native American Land for Data Centers

Big Tech companies are targeting Native American lands for the placement of new data centers.

AI/ML arXiv cs.AI

From Black Box to Executable Logic: Explainable Reinforcement Learning through Prolog Expert Systems

Researchers propose a method to transform deep reinforcement learning policies into readable and executable Prolog programs for better explainability.