September 2025 — page 1 of 2
- New Pruning Technique Boosts Low-Rank Adapters for Efficient LLM Fine-Tuning· AI
- LLMs Tested on Long-Term Business Strategy Through Month-Long Retail Simulator· AI
- How Average Scores Hide AI Model Weaknesses in Popular Benchmarks· AI
- New Benchmark Tests AI's Ability to Solve Frontier Mathematics Problems· AI
- Researchers Use AI Agents to Generate Persuasive Dialogue Training Data at Scale· Agentic AI
- New Framework Tackles Data Noise and Bias in Automated Fact-Checking Systems· AI
- Large Language Models Fall Short in Robotic Planning When Paired With Formal Reasoning· AI
- New method improves AI model adaptation by smarter human annotation selection· Machine Learning
- New method tackles bias in AI medical diagnosis systems using multiple data sources· AI
- New Mathematical Framework Links Neural Networks to Reproducing Kernel Spaces· AI
- New quadratic programming method solves Nash equilibrium in complex multiplayer games· AI
- Researchers develop zero-shot detector for AI-generated Vietnamese text with 98% accuracy· AI
- New AI framework teaches autonomous vehicles to drive for collective traffic efficiency· AI
- Researchers Reveal "Plot Twist" Attack That Jailbreaks Multimodal AI with Visual Narratives· AI Security
- Open-source projects show bursty collaboration patterns driven by social triggers· Software Engineering
- New Underwater AI Benchmark Exposes Gaps in Ocean Robot Intelligence· AI
- New Framework Speeds Up Automated Test Generation for AI Code Models Without Losing Quality· AI
- Evolution Strategies Emerge as Powerful Alternative to Reinforcement Learning for LLM Fine-Tuning· AI
- New framework breaks down language into components for more precise object detection· AI
- First watermarking technique for diffusion language models tackles out-of-order token generation· AI
- Researchers unveil Fast Feature Field, a novel framework for event-based camera processing achieving state-of-the-art vision tasks· AI
- New Framework Combines AI with Simulations to Fix Hallucinations in Scientific Q&A· AI
- New AI Framework Adopts Test-Driven Development for Large Repository Code Generation· Agentic AI
- New Method Lets Code Completion Models Follow Developer Instructions Better· AI
- Theory Behind Neural Network Scaling Gains Ground in Feature Learning· AI
- Vision-Language Models Stumble on Visual Math Problems, Revealing Blind Spot in Reasoning· AI
- New system trims medical AI reasoning overhead by up to 6x while keeping accuracy intact· Agentic AI
- New framework exposes gaps in AI program-repair tools' actual capabilities· AI
- Training Models to Predict When LLMs Are Right, Not Just to Trust Themselves· AI
- Researchers Introduce NeMo: A New Benchmark for Testing Video AI Understanding of Long Temporal Sequences· AI
- New method decodes what's happening inside large language models using symbolic vector algebra· AI
- Hybrid AI model achieves major leap in detecting eye lesions from retinal images· AI
- New Benchmark Reveals Vision-Language Models Struggle With Temporal Medical Reasoning· AI
- New Framework Proposes Treating AI Agents as First-Class Services· Agentic AI
- EasySteer brings production-ready steering to language models with 20x speed boost· AI
- New technique helps diffusion language models improve themselves without extra training· AI
- Researchers map how diffusion models organize image information, revealing the mechanics behind classifier-free guidance· AI
- Language Models Struggle to Update Beliefs as New Evidence Emerges· AI
- New Benchmark Tests AI's Ability to Spot Home Safety Hazards During Autonomous Exploration· AI
- New approach tackles class imbalance in graph neural networks through invariant feature learning· AI
- New System Cuts Training Inefficiencies in Diverse AI Models by Rethinking Pipeline Parallelism· AI
- New Framework Sharpens Image Generation by Anchoring Guidance to Key Semantic Features· AI
- New technique dramatically cuts AI hallucinations by injecting facts into the neural attention mechanism· AI
- New Technique Speeds Up Vision Transformer Model Inversion While Preserving Knowledge Transfer· AI
- New AI method improves spatial understanding in image generation from non-standard perspectives· AI
- Extending game-theoretic LLM alignment to multiple competing policies· AI
- New Framework Tackles Core Efficiency Problem in Mixture-of-Experts Language Models· AI
- New Benchmark Tests LLM Agents on Real-World Software Compilation Tasks· AI
- AI-extracted code features match embeddings while staying human-readable· AI
- New Technique Improves How AI Models Learn User Preferences Through Better Reasoning· AI
- New Algorithm Lets AI Models Learn from User Feedback in Real Time· AI
- New technique makes AI safety guardrails work across multiple languages without retraining· AI
- AI Models Learn Strategic Tool Use Beyond Simple Timing, Boosting Math Problem Accuracy· Agentic AI
- New Diffusion Model Tackles Synthetic Data Generation for Clinical Survival Studies· AI
- QoNext Framework Brings Network Quality Standards to AI Chatbot Evaluation· AI
- DeepTravel brings reinforcement learning to autonomous trip planning· Agentic AI
- New Method Enables AI Policies to Adapt When Training and Real-World Tasks Don't Match· AI
- Study Compares Eight Self-Driving Motion Planners Using Unified CARLA Test Platform· AI
- New AI model converts low-quality cone-beam CT scans into clinical-grade CT images for radiotherapy· AI
- New method clarifies what multimodal AI sees when generating image descriptions· AI
- Teaching Small Language Models Their Own Reasoning Style Boosts Problem-Solving· AI
- Reasoning Distillation vs. Scaling: Study Questions Whether Longer Training Data Pays Off· AI
- Vision-language models tackle medical image analysis using reference images as guides· AI
- New benchmark tests how well AI models generate ML pipelines across languages· AI
- Solution Diversity Emerges as Key Predictor of LLM Problem-Solving Success· AI
- New Algorithm Tackles Bias in Data Discretization to Ensure Fairer Group Representation· AI
- Sleeping Beauty Problem Gets a Betting-Strategy Fix· AI & Philosophy
- Surrogate Model Accuracy Alone Doesn't Guarantee Better Configuration Tuning, New Framework Shows· AI
- New dataset of 1.8M code edits co-authored by AI agents and humans shows promise for training better coding models· AI
- FastLBP: GPU Framework Accelerates Program Analysis Using Belief Propagation· Software Engineering
- LLMs fabricate nonexistent libraries in generated code, creating security and build risks· AI
- FP8 Quantization Offers Practical Efficiency Gains for Large Language Model Training· AI
- Foundation models falter on engineering: New benchmark reveals weak multimodal reasoning· AI
- Language shifts how AI judges morality, but underlying reasoning stays universal· AI
- How Language Models Learn from Examples: Information Filtering as the Core Mechanism· AI Research
- How Users Signal When They Want to Switch Roles with AI Partners· Human-AI Collaboration
- Quokka uses large language models to generate loop invariants and speed up program verification· AI
- New technique enables faster parallel token generation in diffusion language models while maintaining output quality· AI
- New Framework Bridges the Gap Between Knowledge Graphs and LLM Reasoning· AI
- New Benchmark Exposes Vision-Language Models' Weakness in Visual Comparisons· AI
- Smaller Models Can Now Predict Large Language Model Reasoning at Fraction of Cost· Agentic AI
- Sycophancy in AI Isn't One Problem—It's Multiple Independent Behaviors· AI
- New Dataset Enables Detection of AI-Written Text at Character Level· AI
- Advanced AI models gain persuasive power but also become more resistant to influence· AI
- New Dataset Tests AI's Struggle With Complex Math Across Scientific Papers· AI
- New Scheduling Method Slashes Language Model Serving Latency by Over 15x· AI
- New tool boosts detection of hidden code problems in machine learning systems· Machine Learning
- New offline learning technique improves reasoning in large language models with minimal computational cost· Agentic AI
- New Survey Maps How AI-Generated Networks Can Both Attack and Defend Cybersecurity Systems· AI Security
- Researchers Introduce SafeCoDe to Make AI Vision Models Better at Context-Aware Safety Decisions· AI Safety
- New Open Dataset Brings Large-Scale Speech Data to Finnish and Swedish TTS Systems· AI
- New System Lets Users Fix Speaker ID Errors in Meeting Transcripts with Real-Time Feedback· AI
- New method detects LLM hallucinations by analyzing sparse MLP signals without training· AI
- Clotho predicts LLM test failures from hidden states alone, slashing verification costs· AI
- New Technique Boosts Text-to-Image Quality by Rethinking Noise Generation During Inference· AI
- Study clarifies how language models function in open-ended military simulations· AI
- New AI Model Reconstructs Ocean Wave Heights From Limited Buoy Data· AI
- Embedding-Based RAG Beats GraphRAG for Math Textbook Question Answering· AI
- New Study Reveals Gaps in AI's Ability to Track Real-Time Information Trends· AI
- Multi-Agent Language Analysis Helps AI Identify Native Language More Reliably· AI