Other Hacker News

The US ambassador had Belgian police stop our reporting

A report regarding US ambassadors and Belgian police interfering with journalistic reporting.

Other Hacker News

All Logic, No Bite

A Hacker News discussion thread titled 'All Logic, No Bite'.

Tech Business/VC The Verge

Meet the lawyer who beat Elon Musk — twice

A profile of the lawyer Bill Savitt who has successfully litigated against Elon Musk in multiple cases.

AI/ML arXiv cs.AI

A Systems-Level Analysis of Sensitivity, Robustness, and Stability in Retrieval-Augmented Generation

An empirical study analyzing the sensitivity, robustness, and stability of RAG systems, emphasizing the need for multi-stage failure analysis over simple accuracy metrics.

AI/ML arXiv cs.AI

Multi-Agent DRL for QoS and Energy Optimization in RIS-Enabled Open-RAN Industrial 6G TN/NTN Networks

Proposal of a RIS-enabled Open-RAN framework for 6G networks using multi-agent deep reinforcement learning for energy and QoS optimization.

AI/ML arXiv cs.AI

Operating Regimes of Decentralized Learning Under Mobility and Bandwidth Constraints

A study on decentralized learning over wireless systems, identifying operating regimes based on mobility, bandwidth, and contact patterns.

AI/ML arXiv cs.AI

The Crowded Embedding Space: A Mean-Field Mechanism for Emergent Marginalization in Retrieval-Augmented Agents

Theoretical analysis of how high document density in embedding spaces can marginalize minority interests in retrieval-augmented agents.

AI/ML arXiv cs.AI

PIXELRAG: Web Screenshots Beat Text for Retrieval-Augmented Generation

Introduction of PixelRAG, a method that performs RAG using website screenshots instead of parsed text, improving accuracy and efficiency.

AI/ML arXiv cs.AI

Auditing LLM-Governed Social Robots with Culture-Specific Moral Gradients

A gradient-based audit framework to evaluate LLM-governed social robots' moral trade-off behavior across different cultural preferences.

AI/ML arXiv cs.AI

Agentic Safety is an Epistemic Property, Not a Behavioral One

A theoretical argument that AI safety should be viewed as an epistemic property of the learner's teachability rather than just behavioral snapshots.

Other Hacker News

Should every baby's DNA be sequenced?

A Hacker News discussion thread exploring the ethical and societal implications of universal infant DNA sequencing.

Software Engineering Hacker News

Parse, Don't Validate – In a Language That Doesn't Want You To

A technical article discussing the 'Parse, Don't Validate' paradigm in software design and its application in various programming environments.

AI/ML arXiv cs.AI

$M^3 QuestionIng$: Multi-modal Multi-span Medical Question Answering

The paper introduces $M^3QAFrame$, a multi-modal, multi-span medical question-answering framework designed to use both text and images to improve medical query accuracy.

AI/ML arXiv cs.AI

High-Dimensional Concentration and Retrieval Instability in Embedding Spaces: Implications for Retrieval-Augmented Generation

An investigation into how high-dimensional geometry causes retrieval instability and loss of contrast in embedding spaces, impacting RAG systems.

Other arXiv cs.AI

"AI Watermarking": Bridging Policy Discourse and Technical Capabilities

A critical analysis of the gap between US and EU policy discussions regarding AI watermarking and the actual technical capabilities for tracking content.

AI/ML arXiv cs.AI

When Medical Safety Alignment Fails: A Benchmark for Evaluating LLMs on High-Risk Medical Queries

The paper introduces MedHarm, a benchmark for evaluating the medical safety and refusal capabilities of LLMs when facing high-risk queries.

AI/ML arXiv cs.AI

Insidious by Design: Implications of Large Language Model algorithmic bias for the Global South

An exploratory study highlighting how LLM outputs can subtly reproduce Western-centric biases and racial hierarchies, particularly affecting the Global South.

Other arXiv cs.AI

Ground Truths in Suicide Research: The Current State of AI-Based Suicide Detection in Social Media

A research synthesis exploring the limitations and lack of ground truth in AI-based suicide detection from social media data.

AI/ML arXiv cs.AI

LLM-Ideoplasticity: Measuring Ideological Plasticity in the Political Behavior of LLMs as a Context-Conditioned Distribution

The study proposes that LLM political ideology should be viewed as a conditional distribution rather than a fixed point, exhibiting high context sensitivity.

AI/ML arXiv cs.AI

HyBIRD: Hyperbolic Bridge Retrieval and Diagnosis for Methodology Inspiration Retrieval

The paper proposes HyBIRD, a framework for methodology inspiration retrieval using hyperbolic bridge retrieval to improve research paper discovery.