Updated
All the news
Everything the newsroom published, in chronological order. Each item carries origin, sources and reading time.
3Today
4Yesterday
64This week
Thu, May 14
3 stories
RESEARCH Every Guardrail Classifier Tested Fails Formal Safety Verification 
RESEARCH Math Proof Shows Transformer Attention Stabilizes Predictably 
RESEARCH AI Agents Bypass Software Engineering, Risk Production Failure Wed, May 13
8 stories
RESEARCH SLIM improves LLM agent performance 7 percentage points 
RESEARCH Shepherd Raises Agent Accuracy 90% With Forking Traces 
RESEARCH WildClawBench: Claude Opus Clears 62% in Real-World Agent Evaluation 
RESEARCH Sparse MoE Models Match Dense Transformers at 3× Faster Inference 
RESEARCH Muon Optimizer Achieves 2× Speed Over AdamW in Production LLM Training 
RESEARCH CIVeX Logs Zero False Executions in Confounded Workflows 
RESEARCH Paper Dismantles Causal Discovery Claim in Prediction Models 
RESEARCH Frozen Models Encode Semantic Roles Without Fine-Tuning Tue, May 12
7 stories
RESEARCH Flow-OPD Raises Stable Diffusion Accuracy to 92 From 63 
RESEARCH Conformal Path Reasoning cuts knowledge graph answer sets by 40 percent 
RESEARCH Los Alamos Team Trains 8B Model That Generalizes Across Reasoning Benchmarks 
RESEARCH Longer Context Degrades LLM Cooperation, Study Finds 
RESEARCH AutoTTS Cuts Inference Costs 69.5% With Learned Test-Time Scaling 
RESEARCH Coupling Tax: Reasoning Mode Cuts Accuracy Under Token Limits 
RESEARCH Frontier Models Disagree on Ambiguous Policies, DRIP-R Shows Mon, May 11
8 stories
RESEARCH ActCam Controls Video Cameras and Characters Without Fine-Tuning 
RESEARCH Optimizer-Model Consistency Cuts LLM Forgetting in Finetuning 
RESEARCH Math AI Training Solver Accuracy Rises 21.4% With Verifier-Backed Generation 
RESEARCH Arena Analysis: 66% of Leaderboard Votes Cancel Out 
RESEARCH SIRA Outperforms Dense Retrieval Without Training or GPU Infrastructure 
RESEARCH UniPool cuts MoE parameter budget 34 to 58 percent 
RESEARCH DeepMind Math AI Hits 48% on Research-Grade Problems 
RESEARCH Sparse MoEs retain accuracy at 87.5% weight pruning Sun, May 10
7 stories
RESEARCH Rice and Apple researchers cut image-generation FID 22% with token fix 
RESEARCH First-Token Entropy Rivals Multi-Sample Hallucination Detection 
RESEARCH MRI-Eval Finds LLMs Score 97% on Flashcards, 30% on Open Recall 
RESEARCH Q2RL Reaches 100% Success on Peg Insertion, Outpacing BC and IBRL 
RESEARCH Purdue and Georgia Tech Prove Transformers Extract Nonlinear Features in Context 
RESEARCH LongSeeker Beats Competitors on Long-Horizon Tasks 
RESEARCH Verkor's Agentic System Closes RTL-to-Layout in 80 Hours Sat, May 09
2 stories
RESEARCH Multi-Agent LLMs Lose One-Third Quality But Signal Recovery Path 
RESEARCH Claude's Safety Tests Fail When Model Hides Suspicions Inside Fri, May 08
5 stories
RESEARCH Mozilla Found 12 Critical Firefox Bugs Using Claude Mythos AI 
RESEARCH SymptomAI Outperforms Clinicians 2.47x in Real-World Trial 
RESEARCH OpenSeeker-v2 beats Alibaba's Tongyi on agentic search benchmarks 
RESEARCH Automated agent recommender cuts multi-agent system engineering steps to one 
RESEARCH Dreadnode Framework Cuts AI Red Teaming from Weeks to Hours Load more stories Loading…