What would you like to explore?
Browse the full archive →Trending papers
Most discussed on X · last 14 days
The Agentic Scientist: Orchestrating Coding Agents for Automated Scientific (Co-)Discovery
This paper shows that coding agents become more useful research collaborators when an orchestration system helps them test ideas, share findings, and tie claims to evidence—while keeping humans responsible for the final checks.
Scalable-Softmax Is Superior for Attention
This paper introduces Scalable-Softmax, a simple yet effective replacement for traditional Softmax in Transformer models that helps maintain strong attention over longer sequences and improves performance on long-context tasks.
Learning When to Think: Utility-Guided Adaptive Reasoning for Efficient Text Retrieval
This paper introduces AURA, a retrieval system that checks whether each reasoning step improves a query and stops when it doesn’t, delivering better search results with far less time and text.
Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes
This paper turns city-scale 3D object finding from picking among pre-made boxes into an active search guided by landmarks and close-up views, making it more accurate and about ten times faster.
“Two ordinary phrases—“best” and “right now”—made models recommend AI more often, even though users never asked for AI. Without either phrase, 47% of responses included an AI-related recommendation; with both, the rate rose to 73%. “Right now” had the larger effect, increasing the rate by about 18 percentage points on average.”
Where Do LLMs Learn to Recommend AI?Daily Publications
Personal agents spent this window running into other people's systems. The most-shared example was a job interview. A blogger reportedly found an AI recruiter on the video call, connected his own agent and stepped out of frame, leaving the two to keep discussing the vacancy. Two retellings drew 33,8...