Updated
All the news
Everything the newsroom published, in chronological order. Each item carries origin, sources and reading time.
0Today
0Yesterday
25This week
Tue, Aug 11
6 stories
RESEARCH Unified LLM Training Reveals Fundamental Mode Conflict 
RESEARCH MMDiff Lets Engineers Audit and Edit Multimodal Model Features 
RESEARCH ArchAgent v2 beats human champions on three-level cache design RESEARCH GENCO replaces three classical power grid solvers with one neural architecture 
RESEARCH Amazon's Consilience Detects Silent Mode Collapse in Confidence-Based Reasoning 
RESEARCH Taboo Stress-Tests LLMs on Production Constraints Mon, Aug 10
5 stories
RESEARCH Muon Optimizer Collapse After Grokking Threatens Production Training 
RESEARCH SABRE Benchmark Exposes Vision Models' Blindness to Visual Contradictions 
RESEARCH CreativeInstruct Recovers Diversity Without Slowing Inference 
RESEARCH Mankind Labs Cuts Agentic Loop Tokens 17–26% With Reversible Memory Eviction RESEARCH CoinRAG Achieves 5.3% F1 Gain on Multi-Hop RAG with Nugget Caching Sat, Aug 08
1 story
Fri, Aug 07
6 stories
RESEARCH CalibForge Trains Models to 30-Point Gains with Solver-Calibrated Tasks 
RESEARCH AV-AIVAT Cuts Agent Evaluation Cost 74-Fold With Certified Stopping RESEARCH Programmatic Tool Calling Beats JSON on New AI Models RESEARCH NVIDIA Publishes Full Greek RAG Playbook Reaching 2.3× Retrieval Gain 
RESEARCH DeepMind's WeatherNext Cyclones Outpace Hurricane Models by Full Day 
RESEARCH Skill Entropy Lifts Reasoning Model Scores From 34% to 68% Thu, Aug 06
6 stories RESEARCH Agent Harness Outweighs Model Choice in Document Reasoning RESEARCH Reasoning Core Beats Rival Procedural Toolkits on Three Reasoning Benchmarks 
RESEARCH OctoLong improves code agents with cross-repository training 
RESEARCH EvolveNet Evolves Agent Harness Instead of Model Weights 
RESEARCH Chain-of-Thought Works at Tiny Scale, Breaking Compute Link RESEARCH SeGaBench Tests LLMs on Compiler-Blind Code Optimization
RESEARCH OctoLong improves code agents with cross-repository training RESEARCH EvolveNet Evolves Agent Harness Instead of Model Weights RESEARCH Chain-of-Thought Works at Tiny Scale, Breaking Compute Link RESEARCH SeGaBench Tests LLMs on Compiler-Blind Code Optimization Wed, Aug 05
9 stories RESEARCH SocietyBench Counterfactual Events Expose Model Blindspots 
RESEARCH ReflectRL Recovers Reasoning Gains from Failed Expert Rollouts 
RESEARCH Test-time scaling regimes require distinct evaluation profiles RESEARCH ALiBi Models Lose Positional Awareness in Long-Context Inference 
RESEARCH Launch Inputs Matter More Than Model Choice for TUI Testing 
RESEARCH Microsoft Maps Multilingual Agent Planning Failures, Lifts Accuracy 5.6% 
RESEARCH GDPevo benchmark shows enterprise agents gain 8 points from experience 
RESEARCH HazMart Quantifies Faithfulness-Safety Trade-Off in Reasoning Models 
RESEARCH Frontier Models Attempt Cheating in 7–14% of Evaluation Runs
RESEARCH ReflectRL Recovers Reasoning Gains from Failed Expert Rollouts RESEARCH Test-time scaling regimes require distinct evaluation profiles RESEARCH ALiBi Models Lose Positional Awareness in Long-Context Inference RESEARCH Launch Inputs Matter More Than Model Choice for TUI Testing RESEARCH Microsoft Maps Multilingual Agent Planning Failures, Lifts Accuracy 5.6% RESEARCH GDPevo benchmark shows enterprise agents gain 8 points from experience RESEARCH HazMart Quantifies Faithfulness-Safety Trade-Off in Reasoning Models RESEARCH Frontier Models Attempt Cheating in 7–14% of Evaluation Runs Tue, Aug 04
3 stories RESEARCH AURORA-LM Matches Larger Diffusion Models at 1B Parameters 
RESEARCH AtumAI Cuts Datacenter Policy Design from Months to Plain Text 
RESEARCH PRECOG Cuts RAG Latency 4500× on Edge SSM Hardware
RESEARCH AtumAI Cuts Datacenter Policy Design from Months to Plain Text RESEARCH PRECOG Cuts RAG Latency 4500× on Edge SSM Hardware Sat, Aug 01
4 stories
Load more stories Loading…