Updated
All the news
Everything the newsroom published, in chronological order. Each item carries origin, sources and reading time.
0Today
0Yesterday
25This week
Wed, May 06
8 stories
RESEARCH SpecKV Boosts Speculative Decoding Efficiency by 56% RESEARCH Contrastive Learning Backdoor Attacks Show Four Critical Failure Modes 
RESEARCH Diffusion Models Cut Compute on Sparse Data with Selective Processing 
RESEARCH GeoSAE Decodes Brain MRI Models With 97% Cross-Cohort Stability 
RESEARCH SubQ Achieves Frontier Accuracy With Subquadratic Architecture 
RESEARCH Reward Model Accuracy Tops Out at 49% on Real-World Preferences 
RESEARCH AI R&D Self-Improvement Hits 60 Percent Probability by 2028 
RESEARCH LightKV halves vision-token cache size in LVLMs Tue, May 05
4 stories
RESEARCH HyCOP Cuts PDE Solver Error on Out-of-Distribution Problems 
RESEARCH Qwen3-VL Accuracy Gains 4.8 Points With Persistent Visual Memory Module 
RESEARCH RunAgent Enforces Deterministic Execution for LLM Workflows 
RESEARCH Benchmark scores mask LLM failures on multi-step tasks Mon, May 04
9 stories
RESEARCH Columbia Releases Open-Source 100 Hz Tactile Sensor 
RESEARCH Quantum Autoencoders Improve ML Security 68% Over Current Defenses 
RESEARCH IISc's Wavelet Neural Network Solves Industrial Simulation Loss Imbalance 
RESEARCH LLMs Refine Graph Structure in EEG Seizure Detection 
RESEARCH Wolf, Fatkhullin, and He Prove RL Global Optimality Under Safety Constraints 
RESEARCH PhyCo Adds Physics to Video Diffusion Without Simulators 
RESEARCH Intern-Atlas Maps 9.4M Methodological Dependencies in AI Literature 
RESEARCH Microsoft generates 1,000 synthetic computers to train agents 
RESEARCH Strait cuts high-priority GPU deadline violations by 11 points Sun, May 03
10 stories
RESEARCH Models Learn to Hide Capabilities From Reinforcement Learning 
RESEARCH Enterprise AI Silent Failures Dodge Detection, Stanford Study Finds 
RESEARCH New Loss Family Fixes Cold-Start RLVR Fine-Tuning 
RESEARCH Carbon-Taxed Transformers Cut Model Memory 49x Without Retraining 
RESEARCH Standard Safety Evals Miss Misalignment in Production Workloads 
RESEARCH Researchers Steer LLM Emotions Through Final-Phase Feature Control RESEARCH RecursiveMAS cuts multi-agent token usage 34.6% to 75.6% 
RESEARCH DV-World Benchmark: AI Data Visualization Agents Score Below 50% on Production Tasks 
RESEARCH Stanford Chip Cuts Inference Energy to One-Seventieth CPU Cost 
RESEARCH Bender et al. Publish Race and Ethnicity Framework for NLP Research Fri, May 01
1 story
Wed, Apr 29
5 stories
RESEARCH Multi-teacher CoT pooling can be computationally hard, active queries fix it 
RESEARCH Safer-Looking LLM Outputs Miss More Critical Diagnoses, Green Shielding Study Finds 
RESEARCH Persona Collapse Undermines Multi-Agent LLM Simulations Across Ten Models 
RESEARCH FIND-Lab releases AgentWard, a five-layer AI agent security framework 
RESEARCH Anthropic finds Claude does not start safety sabotage but will continue it when primed Tue, Apr 28
3 stories
Load more stories Loading…