Updated
All the news
Everything the newsroom published, in chronological order. Each item carries origin, sources and reading time.
3Today
4Yesterday
64This week
Tue, Jun 02
3 stories
RESEARCH SafeSteer cuts alignment tax by targeting sparse safety tokens 
RESEARCH Output Format Drives Faster Accuracy Loss Than Domain Shift in Multimodal LLMs 
RESEARCH SubFit Maintains 84.6% Accuracy While Pruning LLM Layers at 25% Sparsity Fri, May 29
5 stories
RESEARCH Claude Code Spent 58% of Sessions Optimizing a Broken Architecture 
RESEARCH Robot Manipulation Accuracy Jumps 22.5% With Motion-Aware Encoder 
RESEARCH Linear Inverse Problems Don't Protect Against Diffusion Hallucination 
RESEARCH HullFT Method Cuts Test-Time Finetuning Latency Versus SIFT 
RESEARCH GPIC Open-Source Dataset Displaces ImageNet-1K as Standard Training Corpus Thu, May 28
7 stories
RESEARCH Vision-Language Models Show No Advantage in Text-Only Alignment 
RESEARCH Omega-QVLA Cuts Robot Vision Model Memory by 71% Without Retraining 
RESEARCH Production Hardware Tests Needed Before OFT Replaces LoRA at Scale 
RESEARCH Schema.org Metadata Cuts Agentic Retrieval Errors by Two-Thirds 
RESEARCH Stanford Framework Keeps AI Agents Within Violation Targets 
RESEARCH Bidirectional Evolutionary Search Escapes Autoregressive Limits in Reasoning 
RESEARCH MATCHA Outperforms BERTScore by 20% at Detecting Semantic Contradictions Wed, May 27
3 stories
RESEARCH BRANE Cuts Retrieval Agent Costs by 89% Per Query 
RESEARCH RLHF Training Amplifies Model Bias to 100 Percent 
RESEARCH Meta Shrinks Mixture-of-Experts to Smartphones Without Cloud Offloading Tue, May 26
10 stories
RESEARCH Mistral's 30B mixture-of-depths model remains unconfirmed but would fill a code-stack gap 
RESEARCH ActiveGraph Inverts Agent Architecture, Putting Event Log First 
RESEARCH LoopMDM Cuts Training FLOPs 3.3× by Recycling Transformer Layers 
RESEARCH VeriTrace Improves Research Agents Without Scaling Models 
RESEARCH Claw-Anything Benchmark Sets 34.5% Ceiling for Always-On Agents 
RESEARCH OrpQuant Runs 7B Models on Edge Silicon Without Multipliers 
RESEARCH IBM Framework Classifies Code Changes at 84% Recall 
RESEARCH Self-Generated Replay Cuts Catastrophic Forgetting in Fine-Tuned Models 
RESEARCH Stanford Framework Reveals Hidden Flaws in AI Benchmarks 
RESEARCH MobileGym Solves Mobile-Agent Reproducibility at Scale Mon, May 25
7 stories
RESEARCH Study: AI Narrative Explanations Boost User Trust, Not Accuracy 
RESEARCH Model Scale Fails to Predict Extracted Skill Performance 
RESEARCH Five Bugs Killed agentmemory in Seven Days 
RESEARCH Shannon-Hartley Theorem Explains LLM Quantization Regressions 
RESEARCH Complete-muE Lets Teams Transfer Dense Hyperparameters to MoE 
RESEARCH Microsoft's SkillOpt Lifts Agent Accuracy 24 Points via Automated Skill Refinement 
RESEARCH MemAudit Cuts Memory-Poisoning Attacks to 0% Sat, May 23
1 story
Fri, May 22
4 stories
Load more stories Loading…