Sam's News — tech-research — 2026-08-24¶
Health Technology¶
7 Virginia Tech develops smart patch for automatic naloxone delivery during fentanyl overdose¶
Virginia Tech researchers have created a wearable patch that detects fentanyl in the bloodstream and automatically releases naloxone in response.
- Device smaller than a penny with 121 microneedle tips
- Delivers multiple doses over 24-hour lifespan
- Nearly 70,000 overdose deaths in U.S. in 2025, over half fentanyl-linked
- Currently in testing phase at Virginia Tech
Sources: WSLS Web Search Update to: Virginia Tech Develops Smart Patch for Automatic Naloxone Delivery During Fentanyl Overdose
AI Ethics¶
6.5 AI Ethics Frameworks Fail to Translate Across Global Contexts, Study Finds¶
A study of 14 experts across 10 countries shows that core AI ethics values like fairness and transparency are reinterpreted differently depending on local context, challenging the assumption they are universally operationalizable.
Sources: arXiv — Artificial Intelligence RSS
Machine Learning¶
6.5 New algorithm generates extreme event scenarios without extreme data¶
A novel algorithm enables prediction of unprecedented scenarios that critical infrastructure and supply chains are unprepared for, without requiring extreme historical data.
Sources: MIT — Artificial Intelligence RSS
AI¶
6 Multi-modal traffic sign detection advances autonomous driving safety¶
A multi-modal approach to traffic sign detection improves reliability across regions, viewing angles, and weather conditions for autonomous vehicles.
Sources: arXiv — Computer Vision RSS, arXiv — Robotics RSS
6 Multi-modality interaction framework enhances vision-language-action autonomous driving¶
A collaborative multi-modality interaction method improves vision-language-action models for end-to-end autonomous driving decision-making.
Sources: arXiv — Computer Vision RSS, arXiv — Robotics RSS
AI/ML¶
6 JuryProbe: Consensus-Risk Diagnostic for LLM Judge Panels in Factuality Assessment¶
A diagnostic method that identifies when agreement among multiple reference-free LLM judges may mask shared blindspots in factuality assessment, with routing to grounded verification.
Sources: arXiv — Computation and Language RSS
6 When Failures Propagate: Causal Attribution in Agentic Retrieval-Augmented Generation¶
A framework for diagnosing and attributing failures in multi-hop agentic RAG systems, where errors at early retrieval stages may surface only in final answers while later retrieval can repair trajectories.
Sources: arXiv — Computation and Language RSS
6 AsmEvo: Agentic Assembly-Level Optimization of AMD GPU Kernels¶
An agent-based system that optimizes AMD GPU kernel machine code directly using LLMs with functional equivalence verification when source code is unavailable.
Sources: arXiv — Computation and Language RSS
6 KREL: Automatic Medical Coding via Knowledge-Guided LLM Reasoning¶
An LLM-based system for automatic medical coding that assigns ICD codes to clinical notes by reasoning over clinical evidence with knowledge guidance.
Sources: arXiv — Computation and Language RSS
6 Quantization-Aware Healing: Recovering Performance in Compressed 4-Bit LLMs¶
A method to recover reasoning, mathematics, coding, and long-context performance in large language models that have been both structurally compressed and quantized to 4 bits.
Sources: arXiv — Computation and Language RSS