ai09 Sept 2026, 09:00
Researchers Introduce AhaBench to Test How AI Agents Learn from Experience
AhaBench is a new benchmark designed to evaluate whether language agents can continually learn and adapt from prior experience.
Source: arXiv cs.LG
Latest news
From Kazakhstan to the world — from the world to Kazakhstan
ai09 Sept 2026, 09:00
Preprint Explains Policy Necrosis and Redundant Coverage in Option-Critic
A new research preprint reveals why additional options boost performance in Option-Critic and identifies the policy necrosis phenomenon.
Source: arXiv cs.LG
ai09 Sept 2026, 09:00
Researchers Propose MGHRL for Adaptive Hypergraph Representation Learning
A new framework called MGHRL leverages granular-ball splitting to capture multi-granularity high-order relationships in graphs.
Source: arXiv cs.LG
ai09 Sept 2026, 09:00
Researchers Introduce GAMPO Framework for AI Agent Governance
A new preprint demonstrates that the benefits of governing AI agents depend heavily on model capacity and task-grounded specifications.
Source: arXiv cs.CL
ai09 Sept 2026, 09:00
Researchers Release TamilEOT Dataset and Models for Voice Agents in Tamil Speech
A new preprint introduces TamilEOT, an open dataset and lightweight models for semantic end-of-turn detection in Tamil telephone conversations.
Source: arXiv cs.CL
ai09 Sept 2026, 09:00
Study Shows Complementary Query Rewriting Boosts Strong RAG Baselines
A new research preprint reveals that combining query rewriting methods with cost-aware routing significantly improves retrieval in RAG systems.
Source: arXiv cs.CL
ai09 Sept 2026, 09:00
Study Examines Human Role in AI Agent Skill Maintenance
A new preprint finds that LLM agent skill repositories rely on human-governed, AI-assisted workflows rather than fully autonomous maintenance.
Source: arXiv cs.CL
ai09 Sept 2026, 09:00
Study Isolates Key Drivers in Deep Learning Medical Image Registration
A new preprint demonstrates that regularization losses drive most accuracy and deformation gains in medical image registration with zero inference cost.
Source: arXiv cs.CV
ai09 Sept 2026, 09:00
Survey Explores Adversarial Attacks and Defenses in Diffusion Models
A comprehensive survey systematizes adversarial attacks and defenses across image, video, and 3D diffusion models.
Source: arXiv cs.CV
ai09 Sept 2026, 09:00
Researchers Explore Frequency-Aware Task Weighting for YOLOv8 Autonomous Driving
A new preprint explores Frequency-aware Task Weighting to balance multi-task vision models for autonomous driving using YOLOv8.
Source: arXiv cs.CV
ai09 Sept 2026, 09:00
Study Finds Emergent Goal-Directed Attention in Vision-Language Models
A new preprint demonstrates that general-purpose vision-language models can produce human-aligned, goal-directed spatial attention without gaze supervision.
Source: arXiv cs.CV
startups-world09 Sept 2026, 08:54
Type.com Unveils a Shared Workspace for Teams, Claude, and Codex
Type.com has introduced a collaborative workspace designed for teams working alongside AI models Claude and Codex.
Source: Product Hunt