AI
AI Agent Reliability Changes When Tools Look Like Code
A 14-model study finds that the interface around an AI agent can materially change tool accuracy, scaling behavior,…
Section
AI
Artificial Intelligence (AI)
From the newsroom
Stories
AI
Video AI Can Get the Count Right While Missing the Events
A controlled 2,190-video study shows why correct final counts can conceal missed events, incomplete traces, and fragile temporal…
AI
After AI, Quantum Computing Is Searching for Its iPhone Moment
Quantum computing is not the next AI. It could become the specialized engine that expands what AI and…
AI
OctoLong Trains AI to Follow Code Across Repositories
A new research pipeline tests whether dependency-rich software training data can make long-context AI more reliable across files,…
AI
Voice Transcripts Can Make AI Reasoning Less Reliable
A USC-led study finds that voice-transcription rewrites can reduce language-model accuracy more than ordinary keyboard mistakes, especially on…
AI
AtumAI Turns Datacenter Policy Design Into a Search Problem
A Microsoft Azure and UT Austin research framework uses formal constraints, simulation, and feedback to search for better…
AI
ExtractBench Shows Why AI Document Accuracy Is Not Enough
A new enterprise-document benchmark shows why accurate values alone do not make an AI extraction system dependable: completeness,…
AI
AI Agent Judges Are Too Quick to Declare Success, OSReward Study Finds
A new benchmark finds that AI evaluators often accept failed computer-use agent runs as successes, creating a measurement…
AI
ViewMind3D Tests a Training-Free Route to Robot Spatial Reasoning
A new study tests whether modular visual and spatial scaffolding can help general-purpose AI reason about 3D scenes…