Research

Everything in this section in one place — newest stories first.

LatestAI Agents

A Time-Series Foundation Model Just Learned to Say "No" to a Hallucinating LLM

A neuro-agentic framework combining Gemini 2.5 Flash-Lite with Google's TimesFM prevented 33.3% of simulated cyberattacks on the SWaT industrial water treatment dataset, outperforming LSTM (26.7%) and TCN (13.3%) baselines.

AI AgentsJuly 22, 2026

Researchers Map the Rise of "Skills" as the New Currency of AI-Assisted Software Engineering

A new arXiv study (2607.09065) empirically analyzes how software engineering activities are being packaged into reusable 'skills' across public repositories and marketplaces

ResearchJuly 22, 2026

Researchers Model Quantum Network Security as an Adversarial Game, Then Ask an LLM to Explain the Results

Researchers model quantum-network routing under the E91 protocol as an adversarial bandit game between a defender and an eavesdropper, tested across 50 network topologies.

AI AgentsJuly 22, 2026

SheetMind's Three-Agent System Pushes Spreadsheet Automation Past 50% Accuracy

SheetMind scored 100% execution success and 54.8% functional correctness on the 221-task SheetCopilot Benchmark, beating SheetCopilot's 44.3%.

ResearchJuly 21, 2026

A New Distillation Trick Splits Video Diffusion Models to Make Them Faster Without Losing Quality

Transition Matching Distillation (TMD) is a new framework for compressing video diffusion models into few-step generators without sacrificing quality.

RoboticsJuly 21, 2026

New 4D Radar-Camera Framework Treats Occupancy Prediction as a Memory, Not a Snapshot

4DR360 is a new 4D radar-camera framework for autonomous driving that jointly handles 3D object detection and semantic occupancy prediction, submitted to arXiv on July 10, 2026.

Developer ToolsJuly 21, 2026

Diversity Beats Brute Force: New Pipeline Doubles Success Rate in Automated Code Verification

Diversify2Verify is a staged LLM pipeline for the Why3 verifier that generates diverse recursive and imperative implementations, then applies bounded, verifier-guided repair to boost proof success.

ResearchJuly 20, 2026

Soft Prompts Give Crystal-Predicting AI a Memory Boost, No Retraining Required

A new soft prompt learning framework, accepted at UAI 2026, boosts state-of-the-art GNN accuracy on crystal property prediction by 3-15%

ResearchJuly 20, 2026

A Three-Layer Network That Shows Its Work: Inside SAMPAT's Push for Interpretable AI

SAMPAT is a new three-layer neural architecture detailed in a paper posted to arXiv on July 10, 2026, designed for interpretable machine learning.

ResearchJuly 20, 2026

Under a Minute of Data: New Method Maps Speech AI to Phonological Features Without Training

Researchers from CMU, UT Austin, and the University of Tokyo introduced SPAM, a method mapping self-supervised speech model representations to phonological feature activations.

ResearchJuly 17, 2026

EZSMTV3 Brings SMT Muscle to Answer Set Programming's Hardest Constraints

EZSMTV3, developed by Yuliya Lierler and collaborators, extends the EZSMT+ CASP framework with a more expressive input language.

LLMsJuly 17, 2026

Vision-Language Models Keep Failing to See the World From Someone Else's Eyes

FlipSet, a new perspective-taking benchmark, found that most of 103 tested VLMs scored below chance, with ~75% of errors reflecting the model's own camera viewpoint rather than the rotated one; it won Best Undergraduate Student Paper at CogSci 2026

ResearchJuly 17, 2026

New Study Shows Building Feature Reconstruction Can Slash Load-Forecast Uncertainty by Over 100%

A new arXiv paper (July 14, 2026) compares post-hoc and in-model uncertainty methods for smart-building load forecasting.

ResearchJuly 17, 2026

New Saliency-Guided Diffusion Framework Aims to Fix Medical Imaging's Data Scarcity Problem

MedDiffuseMix is a saliency-guided diffusion framework designed to generate synthetic training data for medical imaging classifiers without distorting diagnostic regions.

ResearchJuly 17, 2026

New Benchmark Shows ITeM-HM Triples Success Rate for Migrating Android Tests to OpenHarmony

ITeM-HM hit an 81% success rate migrating Android UI tests to OpenHarmony, a 214% relative gain over the original ITeM method's 26%.

AI AgentsJuly 17, 2026

GroundShot Tackles AI Video's Consistency Problem Without Retraining a Single Model

GroundShot is a new training-free, model-agnostic framework designed to fix entity drift in AI-generated multi-shot video.

ResearchJuly 17, 2026

New Paper Extends Kolmogorov Complexity Theory to Symmetry Groups with a "Geometric Coding Theorem"

A new arXiv paper, 'CAS I: A Geometric Coding Theorem,' launches a series on Computational Algorithmic Statistics, extending algorithmic information theory to symmetry groups.

ResearchJuly 17, 2026

Machine Learning Model Estimates "Brain Vascular Age" from Ultrasound Signals, Flags Disease-Linked Aging

A machine learning model estimated cerebrovascular age from Transcranial Doppler ultrasound of the middle cerebral artery in 445 subjects (168 healthy, 277 diseased).

All articles