SourceJun 20, 20262026-06-203 minVerified
How LLMs Encode Essay Quality for Scoring
Examines how LLMs encode essay quality in hidden representations and whether those signals persist across prompt changes.
SourceJun 20, 20262026-06-203 minVerified
Interpreting Style-Caption TTS With Cross-Attention Attribution
A look at why cross-attention attribution matters for interpreting word-level style control in caption-based TTS.
SourceJun 20, 20262026-06-204 minVerified
JustDiag for Auditable Root Cause Analysis in LLM Workflows
Why JustDiag reframes LLM root cause analysis around evidence, alternatives, contradictions, and uncertainty.
SourceJun 20, 20262026-06-204 minVerified
Measuring False Intervention in DeFi Supervisory AI Agents
Why DeFi supervisory AI should measure false intervention separately from accuracy, with practical checks for evaluation.
SourceJun 20, 20262026-06-204 minVerified
Query Placement Matters in Diffusion LLM In-Context Learning
Why query placement may affect diffusion LLM in-context learning, and what prior position-bias results imply.
SourceJun 20, 20262026-06-204 minVerified
When Learner-Based Drift Detection Works Better in Streaming
Explains when learner-based drift detection outperforms statistical tests in streaming ML and what matters operationally.
SourceJun 19, 20262026-06-194 minVerified
Can General Models Extract Legal Networks Reliably
Using FineREX, this examines why legal-record extraction for smuggling knowledge graphs needs domain-specific schemas and review.
SourceJun 19, 20262026-06-193 minVerified
Self-Review Alignment for Safer LLM Reasoning Outputs
Explores combining a conscience step with DPO so LLMs review reasoning during inference while balancing safety and performance.
SourceJun 12, 20262026-06-124 minVerified
Reframing Shielded RL as Design-Time Structure Analysis
A concise look at shielded RL reinterpreted as a design-time tool for structural safety analysis, not runtime blocking.
CommunityJun 12, 20262026-06-123 minVerified
Rethinking AI Job Impacts Beyond Mass Unemployment Fears
Official reports suggest AI is reshaping tasks and productivity before causing broad job losses.
SourceJun 12, 20262026-06-123 minVerified
Rethinking AI Loss of Control Through Operational Definitions
Examines vague AI loss-of-control language and reframes it around goals, audits, interruption, and rollback.
SourceJun 2, 20262026-06-024 minVerified
Why Mechanistic Interpretability Needs Auditable Validation Rules
Mechanistic interpretability matters, but auditable, reproducible validation rules are what safety-critical AI needs.
SourceJun 2, 20262026-06-024 minVerified
TriLens Tracks Hallucination Signals Across LLM Internal Layers
TriLens explores white-box hallucination detection by tracking layer-wise entropy signals before incorrect answers emerge.
SourceJun 1, 20262026-06-015 minVerified
Do Warm Personalized AI Replies Persuade Users More?
Examines how contextual personalization and warmth affect trust, persuasion, and reliance in conversational AI.
CommunityJun 1, 20262026-06-014 minVerified
Why AI Stops Reproducing Lyrics and Long Texts
Why AI services often block long copyrighted text reproduction but allow transforms of user-provided text.
SourceJun 1, 20262026-06-015 minVerified
Why Linear Recurrent Memory Works in POMDP RL
A look at why linear recurrent memory can work in partially observable RL through an HMM belief filtering view.
SourceMay 30, 20262026-05-304 minVerified
Expert-Guided LLMs for Marine Lead Data Extraction
How expert-guided LLM agents structure marine lead and isotope data hidden in scientific literature.
CommunityMay 29, 20262026-05-294 minVerified
Coding Models Differ in Execution and Planning Styles
Coding model differences appear not in prose quality but in planning, tool use, and context handling scope.
SourceMay 29, 20262026-05-295 minVerified
Measuring Neural Networks' Preference for Simpler Solutions
A look at a proposed metric that approximates neural simplicity bias with data-dependent polynomials and its limits.
SourceMay 29, 20262026-05-294 minVerified
Q-Guided Alignment for Return-Conditioned Offline RL Control
Examines limits of RTG-only conditioning and how Q-guided alignment aims to improve controllability and reliability in offline RL.
CommunityMay 29, 20262026-05-295 minVerified
Reading AI Pricing Through Limits and Infrastructure Costs
AI pricing is better understood through usage caps, fallback rules, and inference infrastructure efficiency, not subscription fees alone.
SourceMay 28, 20262026-05-284 minVerified
From Black-Box Grading to Rubric-Based Explainable Scoring
A look at rubric- and concept-based grading that makes open-ended scoring more reviewable, editable, and accountable.
SourceMay 28, 20262026-05-283 minVerified
Evaluating AI Agents for E-Commerce Dispute Resolution Tasks
CyberJurors evaluates agent systems on multi-round, multimodal evidence handling and platform rule adaptation in e-commerce disputes.
CommunityMay 28, 20262026-05-284 minVerified
How Far Can Multimodal AI Be Trusted
Why multimodal AI still struggles with charts and scientific figures, and how to verify image-based conclusions in practice.