§ BEAT
Research
Why Production Agents Fail Without Harness Infrastructure
KV-Fold Extends Transformer Context to 128K Without Retraining
27M Attractor Model Beats GPT o3 on Logic Puzzles
Sparse-to-Dense RL Lifts MATH Scores to 78.5% on Small Models
Standard load-balancing losses degrade SMoE expert specialization by 3x
VECA Cuts Vision Transformer Inference Cost to Linear Time
Los Alamos Team Trains 8B Model That Generalizes Across Reasoning Benchmarks
AutoTTS Cuts Inference Costs 69.5% With Learned Test-Time Scaling
ActCam Controls Video Cameras and Characters Without Fine-Tuning
SIRA Outperforms Dense Retrieval Without Training or GPU Infrastructure
UniPool cuts MoE parameter budget 34 to 58 percent
DeepMind Math AI Hits 48% on Research-Grade Problems
Verkor's Agentic System Closes RTL-to-Layout in 80 Hours
Mozilla Found 12 Critical Firefox Bugs Using Claude Mythos AI
SymptomAI Outperforms Clinicians 2.47x in Real-World Trial
SpecKV Boosts Speculative Decoding Efficiency by 56%
SubQ Achieves Frontier Accuracy With Subquadratic Architecture
AI R&D Self-Improvement Hits 60 Percent Probability by 2028
PhyCo Adds Physics to Video Diffusion Without Simulators
New Loss Family Fixes Cold-Start RLVR Fine-Tuning
RecursiveMAS cuts multi-agent token usage 34.6% to 75.6%
ElementsClaw Screens 2.4 Million Crystals in 28 GPU Hours, Finds Four New Superconductors
42-Author arXiv Survey Defines Three Levels for Agentic World Models
David Silver's Ineffable Intelligence Raises $1.1B to Replace Human Training Data
Agentic Framework Hits 83% Intent Accuracy by Confining LLM to Query Parsing
OpenAI Folds Codex Into GPT-5.5, Forcing Enterprise Migration at 20% Price Hike
DeepMind Aletheia Solves 6 of 10 Research Math Problems, Refuses to Fake the Others
