New Technology– category –
-
LLM Agent Design: Deep learning approach for 11% gain (2026)
arXiv Overview A new arXiv research proposes constructing Large Language Model (LLM) agent systems in a modular manner, drawing analogies to deep learning modules. By treating prompts as weights and prompt optimization as back-propagatio... -
MLE-InsightForge: Agentic ML system and 123.9% Kaggle boost
arXiv Overview New research introduces 'MLE-InsightForge,' an agentic Machine Learning Engineering (MLE) system that achieved a remarkable 123.9% improvement in normalized private-test scores on Kaggle competitions over a no-insight base... -
Osaka University: AI-nanotech method for space life detection
EurekAlert! USA Overview Researchers at Osaka University have developed an electrical method combining gold nanogap electrodes with machine learning to distinguish mirror-image amino acids, a critical step in detecting signs of life in s... -
LLM-IDEA: Autonomous scientific discovery and model specs
arXiv Unknown Overview LLM-IDEA, an identifiability-driven experimental agent, has been developed to autonomously discover mechanistic world models, addressing the critical and often overlooked challenge of identifiability in AI-driven s... -
AO-DA: Llama-3.1-8B edge agent robustness and failure rates
arXiv USA Overview This research introduces a Decoupling Architecture (AO-DA) for edge LLM agents, separating logical inference from persona expression to significantly mitigate 'persona-logic interference' caused by conversational histo... -
MTOR explained: AI video detection specs and 97.76% AP rate
arXiv USA Overview This paper introduces MTOR, a multimodal AI-generated video detector that leverages video-level multimodal semantics and a novel concept of Temporal Over-Regularity (TOR) to identify AI-generated videos with high preci... -
Google Launches Trillium TPU-Equipped Satellite on October 1st for ‘Project Suncatcher’ to Test Spaceborne AI Data Centers
GIGAZINE Japan Overview Google announced on September 24, 2026, the October 1st launch of a prototype satellite carrying its sixth-generation Trillium TPU as part of 'Project Suncatcher.' This experiment aims to evaluate AI chip function... -
Infineon Ships Samples of 120-Amp Power Chips for AI Accelerators, Enhancing Vertical Power Delivery and Liquid Cooling Integration
The Elec Inc. South Korea Overview Infineon Technologies has begun sampling 120-Amp smart power stage (SPS) chips for AI accelerators. These chips achieve an industry-leading power density of 2A/mm² in a compact 6x6x0.8mm package, featur... -
CompMat-Bench: Benchmarking AI Agents for Computational Materials Science Announced on arXiv, Accelerating LLM-Driven Materials Research with Pass Rate Metrics
arXiv USA Overview arXiv has announced "CompMat-Bench," the first benchmark evaluating AI agents' effectiveness in computational materials science tasks. This benchmark measures the "pass rate" of Large Language Model (LLM)-based agents ... -
UTSA Unveils ‘Genesis’ AI Chip: Achieves On-Device Continual Learning, Solving Catastrophic Forgetting with 30-100x Energy Efficiency
UT San Antonio Today USA Overview Researchers at the University of Texas at San Antonio (UTSA) have unveiled 'Genesis,' a groundbreaking AI chip that avoids forgetting old information while learning new data. This neuromorphic accelerato...
