June 2025 — page 1 of 2
- Training Cybersecurity AI on a Fraction of the Data Typically Required· AI
- Sparse autoencoders excel at uncovering hidden concepts, not applying known ones· AI
- New Lightweight Method Improves Confidence Calibration in Neural Networks· Machine Learning
- Researchers develop new metric to measure safety margins in AI alignment· AI Safety
- New computational framework unifies transportation network modeling across forecasting, optimization, and learning· AI
- Comprehensive survey maps how LLMs are transforming automated code repair· AI
- Bisection-based method cuts effort to identify duplicate compiler bugs by a third· Software Engineering
- Training LLMs to Spot and Fix Their Own Security Gaps· AI
- Researchers align LiDAR and camera data in single frame for autonomous driving· AI
- Study Compares AI Counseling Models in Japanese: Structured Prompts Outperform Minimal Instructions· AI
- New benchmark reveals how LLMs encode hidden preferences when making moral decisions· AI
- Spherical Cauchy VAEs Offer New Way to Model Heavy-Tailed Posteriors on Spheres· AI
- New recurrent AI model solves complex reasoning with tiny dataset and no pre-training· Agentic AI
- New Benchmark Tackles Probabilistic Reasoning Over Tabular Data with Bayesian Networks· AI
- Academic Paper Proposes Reimagined Peer Review Built on Blockchain and Open Identity· AI
- RedCoder: AI system automates security testing of code-generating language models· AI Security
- AlphaGenome: DeepMind's AI Tool Decodes How DNA Variants Affect Gene Regulation· AI
- New Framework Pairs Language Models with Spatio-Temporal AI for Complex Multi-Task Reasoning· AI
- New Method Closes Privacy Gaps in Machine Learning Model Unlearning· AI
- Audio-JEPA: Efficient Self-Supervised Learning Framework Predicts Masked Sound Representations· AI
- Researchers develop discrete concept framework to interpret neural networks more transparently· Agentic AI
- Researchers release largest real-world image deblurring dataset from smartphone slow-motion video· AI
- LLM Social Simulations Need Clear Boundaries to Drive Meaningful Research· AI
- New Method Uses Semantic Labels to Refine Causal Discovery Beyond Data Limitations· AI
- Transferable AI Model Tackles Quantum Chemistry's Toughest Problem: Bond Breaking· AI
- LLMs Enable Fault Detection in Test Code Without Running Tests· AI
- New Benchmark Reveals Large Language Models Struggle With Table Reasoning Tasks· AI
- Withdrawn Study Proposes Dynamic Control Method to Reduce LLM Reasoning Overhead· AI
- New Framework Brings Mathematical Guarantees to Quantized Neural Network Training· AI
- New attention mechanism improves hyperspectral image analysis for autonomous vehicles· AI
- Benchmark Scores Don't Tell the Whole Story: New Framework Reveals Hidden Gaps in Vision-Language Model Performance· AI
- Aligned Language Models Narrow Their Output Diversity Through Probability Concentration· AI
- New Framework Combines Weak AI Verifiers to Match Top-Tier Reasoning Performance· Agentic AI
- Withdrawn GUI Agent Framework Proposed Dual-System Reasoning for Automation Tasks· AI
- New LLM framework detects vulnerable APIs in large Java codebases with 95% recall· Security
- Psychologists Need Stricter Standards for Testing Large Language Models as Research Tools· AI
- Language Models Show Promise for Identifying Code Authors, Study Finds· AI
- Researchers Map LLM Adoption Across Writing and Science, Raising Equity Concerns Over AI Detection· AI
- New Dataset Teaches AI to Reason Through Street View Images Like Humans Do· AI
- LLMs Show Troubling Criminal Behavior in Realistic Scenarios, Study Finds· AI Safety
- Study maps user challenges and wins with AI-powered mobile apps· AI
- New AI system detects elderly falls in bathrooms without cameras or wearables· AI
- New framework examines AI systems as hidden cognitive infrastructure reshaping human thought· AI
- New system improves automated speaking tests by understanding images and grammar errors in detail· AI
- LLMs Trained to Recommend Your Next Location With Semantic Awareness· AI
- Generalist AI Models Match Specialist Medical Vision Systems, Often at Lower Cost· AI
- New Tokenizer Slashes Token Counts by Up to 25% While Keeping LLM Performance Intact· AI
- GenRecal: Efficient Knowledge Distillation Method Adapts Large Vision-Language Models to Smaller Devices· Agentic AI
- LLM-Powered Swarms Show Promise but Face Severe Computational Barriers· AI
- New Benchmark Reveals AI Agents Struggle with Form-Filling, Even with Visual Understanding· Agentic AI
- LLMs Show Strong Order Biases in High-Stakes Decisions, Picking Inferior Options· AI
- Survey Maps the Landscape of AI-Generated Sports Commentary· Agentic AI
- New AI Framework Helps Robots Adapt to Hidden, Unpredictable Delays· Agentic AI
- New Technique Cuts AI Image-Misinterpretation Errors by Up to 38% Without Retraining· AI
- New Framework SynBoost Cuts Diffusion Model Inference Time Without Sacrificing Quality· AI
- LLMs boost fixes for regression bugs 1.6x when given original code changes· AI
- New AI Debate Protocol Tackles Honesty Problem in Competitive Reasoning· Agentic AI
- New Training Method Teaches AI Models to Explore Multiple Plausible Next Words· AI
- Survey Reveals How AI Reasoning Techniques Transform Code Generation and Software Engineering· AI
- Finite Memory Q-Function Approach Bridges Theory and Practice in Reinforcement Learning· Machine Learning
- New AI Method Generates Realistic Synthetic Cell Data While Resisting Noise· AI
- How humans read AI learning—new research reveals what people actually understand about RL agents· AI
- New multi-scale fusion technique boosts small object detection in drone imagery· AI
- New Dataset and AI Framework Tackle Hidden Harms in Memes Through Reasoning-Based Detection· AI
- Researchers teach humanoid robot to master kung fu and dynamic movement· Robotics
- New VLM-Guided Framework Uses Rendered Robot Motions to Improve Real-World Manipulation Tasks· Robotics
- New method improves AI steering by better identifying and calibrating latent feature controls· AI
- SheetMind: New LLM Framework Achieves 55% Success Rate on Spreadsheet Automation Tasks· AI
- New AI Framework Learns Human-Like Driving Behavior at Chaotic Intersections· AI
- New system tackles Hebrew's phonetic ambiguity to improve speech synthesis· AI
- Embedding models outweigh chunking strategy in chemistry-focused document retrieval· AI
- New Benchmark Reveals That Advanced AI Models Develop Human-Like Thinking Patterns· AI
- New Framework Cuts LLM Inference Costs by Pairing Small and Large Models with Planning· AI
- New Approach Enables Robots to Map and Recognize Unmarked Objects in 3D Space· AI & Robotics
- Google DeepMind's Veo AI powers Darren Aronofsky-backed short film "ANCESTRA" at Tribeca· AI
- Quasar: A New Programming Language Designed for LLM Code Actions· AI
- Training Language Models Smarter: Stacking Layers Gradually Cuts Compute Costs· AI
- Mamba Architecture Cuts Video Generation Costs by 45%, Offering Efficient Alternative to Transformers· AI
- New Method Improves Text Classification for Low-Resource Languages With Just a Few Labels· AI
- New Framework Splits Robotic Vision and Planning to Boost Manipulation Tasks· Robotics
- Multi-Dimensional LLM Judges Outperform Simple Models in Evaluating Mathematical Formalization· AI
- New Benchmark Tests How Well AI Models Handle Software Engineering Tasks Beyond Code· AI
- Diffusion Models Generate Hard Training Samples Without External Data· AI
- How Diffusion Models are Reshaping Anomaly Detection Across Industries· AI
- New Method Tunes AI to Improve Binary Decompilation While Protecting Accuracy· AI
- MetaTT: A Tensor-Train Adapter That Dramatically Shrinks Fine-Tuning Parameters· AI
- Researchers Learn Environmental Structure by Measuring Minimum Action Distance Between States· Machine Learning
- Framework Locates Key Circuits in AI Models for Safer Steering Without Parameter Changes· AI
- Study reveals how AI bug-fixing agents fail on complex code repairs· AI
- Vision Transformers Trained on Images Now Outperform Time Series Models· Agentic AI
- New transformer architecture enables robots to control unfamiliar bodies without retraining· AI
- New Safety Approach Outperforms Standard Alignment When LLMs Face Fine-Tuning Attacks· AI Safety
- New Self-Play Algorithm Lets Language Models Attack and Defend Themselves to Improve Safety· AI
- New Framework Merges Knowledge Graphs with LLMs to Improve AI-Powered Recommendations· AI
- New AI framework ensures sign language videos match word order and meaning· AI
- New Training Method Boosts Sparse Autoencoders for Instruction-Tuned Language Models· AI Research
- Researchers create vast training dataset for AI to tackle scientific discovery tasks· AI
- New Q-Learning Algorithms Achieve Robust Long-Horizon Decision-Making Under Model Uncertainty· AI
- New Medical AI Model Tackles Hallucinations and Complex Reasoning in Healthcare· AI
- New Training Method Cuts LLM Hallucinations Below 5% by Teaching Models to Admit Uncertainty· AI