February 2026 — page 2 of 2
- New automata-based method scales reinforcement learning to complex formal specifications· AI
- RAG Retrieval Methods Outperform LLMs on Enterprise SQL and API Code Generation Tasks· AI
- New Hourglass Architecture Makes Transformer Language Models More Efficient· AI
- LLMs Could Better Personalize Responses by Understanding Your Personality Traits· AI
- AI System Revives Broken Machine Learning Notebooks by Modernizing Outdated Code· AI
- Training LLMs on Multiple Goals Often Backfires—New Method Could Fix It· AI
- New Method Makes AI Judges Fairer, More Consistent, and Harder to Fool· AI
- Framework lets AI models improve themselves during testing by reflecting on failures· AI
- New Framework Makes System Reliability Decisions Traceable Through Code· Software Engineering
- Uber's AI-powered chaos testing finds critical mobile app failures at scale· AI
- New Framework Boosts SQL Query Generation from Natural Language with Lightweight Models· AI
- TIDE: Self-Optimizing Inference Engine Speeds Up Large Language Models· AI
- New Platform Automates Linguistic Annotation Tasks Using AI Models· AI
- New AI Framework Uses Visual Information to Fix Code Bugs More Effectively· AI
- New Tool Uses AI to Automatically Verify Code Changes and Catch Hidden Bugs· AI
- OmniMoE Scales Micro-Specialized Experts to Boost Language Model Efficiency· AI
- New Method Lets Smaller AI Models Learn From Larger Ones Without High Costs· AI
- Orthogonal Features Could Make AI Interventions Safer and More Predictable· AI
- New Method Moves AI Safety Controls Offline With Surgical Weight Edits· AI Safety
- Withdrawn Paper Proposes Training-Free Steering Method to Fix Tool-Calling Brittleness in Language Models· Agentic AI
- New Method Cuts AI Hallucinations in Medical Imaging Reports· AI
- New Architecture Decouples AI Explanations from Model Inference, Cutting Edge Device Latency by 38%· AI
- LLM surprisal alone doesn't explain how readers stumble on garden-path sentences· AI & Linguistics
- New method uses frozen embeddings to speed up AI planning by 1000x· AI
- Lightweight Student Models Distill Language Model Uncertainty in Real Time· AI
- Researchers Show How Data Structure Explains Mystery Behaviors Across Language Models· AI
- Researchers improve LLM-based exploit generation with new validation pipeline for Java vulnerabilities· AI Security
- How Language and Speech Models Mirror Brain Activity Through Semantic Abstraction· AI
- LLM agent performance isn't a single metric—it depends on what you're actually testing· Agentic AI
- When AI Gets Good at Tasks, Robustness Follows Automatically· Machine Learning
- Declarative Design Alone Won't Cut It: Study Shows Convention Alignment Drives AI-Friendly Code Frameworks· Agentic AI
- H Company launches Holo2-235B-A22B, breaking UI localization records· Agentic AI
- Developers' Security Notes Could Fill Gaps in Automated Code Analysis Tools· Software Engineering & Security
- Information-Theoretic Constraint Revealed for Context Handling in Cognitive Models· AI
- New speech dataset aims to fill conversational gaps in audio AI models· AI
- Researchers Scale Software Engineering Test Environments to Over 800,000 Real-World Tasks· AI
- Study Shows How to Realign Language Model Benchmarks with Real-World Performance· AI
- Researchers map the hidden rules governing how hard to push AI steering interventions· AI
- New Benchmark Reveals Unified Multimodal Models Lack True Integration Between Understanding and Generation· AI Research
- New Framework Trains AI Systems to Verify Mathematical Proofs at Scale· AI
- Rendering Code as Images Could Cut AI Model Costs by Up to 8x, Research Shows· AI
- New Lightweight Method Detects AI Hallucinations Using Only Token Probabilities· AI
- New ROS2 Framework Tackles Multi-Robot Control and Simulation Accuracy for Dual-Arm Robots· Robotics
- Researchers find language models solve problems through vector algebra in hidden subspaces· AI
- LLM Judges Show Consistent Position Bias When Scoring by Rubric· AI
- New framework reveals which features matter most in multi-agent code generation systems· AI
- AI agents now evaluate research code with 71% accuracy, easing peer review bottleneck· Agentic AI
- LLM refusal works through multiple directions, not a single control mechanism· AI Safety
- How the Way You Ask Changes What LLMs Do: Researchers Map Pragmatic Influence on AI Behavior· AI
- New Methods Boost Essay-Scoring AI With Limited Training Data· AI
- New method lets AI models selectively ignore or focus on different input features via natural language· AI
- New fine-tuning method preserves exploration in reasoning models while maintaining performance· AI
- New Framework Improves LLM Understanding of Complex Tables with Hierarchical Structure· AI
- Kimi K2.5 Brings Visual Reasoning to Agentic AI with Coordinated Multi-Agent Framework· Agentic AI
- New Training Method Guides LLM Reasoning Through Step-by-Step Rewards· Agentic AI
- New Framework Dynamically Allocates Computing Power During Inference for Faster AI Reasoning· AI
- New Dataset Trains AI Terminal Agents with 50K Verified Docker-Based Trajectories· Agentic AI
- New framework bridges vision-language models and autonomous driving for rare, high-risk scenarios· AI
- MedBeads: A New Data Structure for Grounding AI Models in Complete Patient Records· AI