March 2025
- Researchers uncover new jailbreak attack targeting LLMs through structured output grammar· AI Security
- Survey maps the trade-offs between reasoning depth and computational cost in large language models· AI
- New Method Learns Physics Simulations Directly from Regular Video· AI
- Study Reveals When Language Models Lie on Their Own· AI
- New survey maps defenses against unauthorized model training using unlearnable data techniques· AI
- Researchers Develop Watermarking System to Detect AI-Generated Imitations of Creative Writing· AI
- SPFinder tackles long patch contexts to improve vulnerability fix detection at scale· Cybersecurity
- New Defense Method Achieves Both High Accuracy and Certified Robustness for Graph Neural Networks· AI Safety & Robustness
- New multi-agent framework helps robots adapt on the fly during complex tasks· Robotics
- Simpler methods can outpace complex AI in software engineering optimization· Software Engineering
- New Bayesian Approach Unlocks More Efficient Robot Learning for Complex Multi-Step Tasks· Robotics & AI
- New Framework Lets Robots Learn Complex Tasks Through Generated Code Plans· Agentic AI
- New mask-based fine-tuning method improves LLM performance without changing weights· AI
- New Open-Source Biomedical Dataset With 24 Million Images Aims to Boost AI Research· AI
- Multi-agent AI system shows promise for analyzing clinical interviews with minimal human oversight· AI
- Researchers Propose New Personality Test Better Suited to How AI Models Actually Behave· AI
- New AI framework generates physically realistic car crash videos from accident data· AI
- New AI Method Uses Formal Logic Ideas to Verify Code Against Natural Language Requirements· AI
- New Benchmark Exposes Major Gaps in AI Vision Models' Spatial Reasoning· AI
- LMs can evaluate other LMs better when given more thinking time, study shows· AI
- Multi-view deep learning approach improves breast cancer detection rates in mammography· AI
- New Benchmark Reveals OCR as Key to Fixing Multilingual Vision-Language Model Gaps· AI
- New LLM Workflow Significantly Improves Modelica Code Generation for Physics Simulations· AI
- Machine Learning Method Uses AI to Verify Physical Laws in Stellar Data· AI
- New AI Framework Tackles Multiple Image Flaws at Once with Adaptive Routing· AI
- LLMs Show Bias Toward Popular Libraries and Python, Often Ignoring Task-Specific Needs· AI
- How Well Do AI Judges Evaluate Summaries in Other Languages? New Multilingual Study Reveals Significant Gaps· AI
- Researchers Use Soft Prompts to Adapt AI Models to Different Cultural Values· AI
- New AI system generates location-specific explanations for deepfake detection· AI
- Researchers replace random masking with noise-shaped patterns in self-supervised video and audio learning· Machine Learning
- New method captures individual motor quirks using AI to power realistic digital avatars· AI
- Withdrawn Study Proposes Method to Remove Training Data from Language Models· AI
- New technique constrains diffusion models to match training data, boosting robotic reliability· AI
- Withdrawn Paper Proposes Two-Stage Training to Boost Language Model Robustness and Generalization· AI
- New Benchmark Tests Whether AI Models Can Write Code Meeting Specific Performance Requirements· AI
- Google's LLM-Powered Code Optimizer Saves Hundreds of Thousands of CPU Cores at Scale· AI
- Deep Networks Unlock Self-Supervised RL: 1000-Layer Models Show Dramatic Performance Gains· AI
- New benchmark tests robot navigation in crowded spaces with social awareness· AI
- New Vision-Language Model Boosts Autonomous Traffic Understanding and Simulation· AI
- New Defense Framework Detects Multi-turn Manipulation Attacks on AI Chat Systems· AI
- AI systems approaching month-long task automation in five years, new benchmark suggests· AI
- Survey maps the landscape of multi-agent AI systems working in tandem· Agentic AI
- Larger language models produce more truthful explanations of their decisions, study finds· AI
- New synthetic data framework improves personalization of vision-language models· AI
- New Method Corrects Biased Sequential Sampling to Match Target Probability Distributions· Machine Learning
- New prompting method helps AI models write better code by breaking down problems hierarchically· Agentic AI
- New Defense System Detects Model Theft Through User Behavior Patterns· AI Security
- New Neural Model Mimics How Humans Anticipate Musical Rhythm· AI & Neuroscience
- Debate Between AI Agents Boosts Code-Solving Ability, New Study Finds· AI
- New AI Framework Brings Fine-Grained Artistic Control to Image Generation· AI
- New Framework Boosts AI Agents' Ability to Navigate Complex Spaces Following Language Instructions· AI
- New AI Model Predicts Human Movement 100× Faster Using One-Step Flow Matching· AI
- New Benchmark and Multi-Agent System Advance AI Reasoning for Surgical Video Analysis· AI
- G-Boost: New Framework Lets Weak Edge AI Models Tap Into Powerful Cloud AI to Boost Performance· AI
- LLMs Offer Alternative to Traditional Programming Through Sampling-Based Computation· AI
- New method breaks AI bias in image recognition by shuffling objects and backgrounds· AI
- New framework teaches AI code generators to follow quality standards alongside correctness· AI
- New Framework Uses Counterfactual Reasoning to Curb Popularity Bias in AI-Powered Recommendations· AI
- New approach uses multiple simulators together to find more reliable autonomous driving bugs· AI
- New algorithm maps protein mutations to understand how language models parse biological sequences· Computational Biology
- New Tool Reveals How Multimodal AI Models Fall for Visual Red Herrings· AI
- Survey Maps the Landscape of Knowledge-Driven Retrieval-Augmented Generation· AI
- Optimal Transport Theory Offers Fresh Foundation for Inferring Causation from Observational Data· AI & Statistics
- New AI System Transforms Binary Code Into Readable Source Code More Accurately· AI
- New Approach Strips Noise from Transformer Time Series Models by Pruning Redundant Dependencies· AI
- New Framework Targets Early Training Phases to Improve Federated Learning with Smart Incentives· AI
- New Framework Tackles "Lost-in-the-Middle" Problem When Large Language Models Generate Long Documents· AI
- New computational method uses information theory to spot repetitive patterns in ancient texts· AI
- New Framework Revitalizes Code Benchmarks to Counter AI Training Contamination· AI
- Graph Neural Networks Guide New Algorithm for Faster Vertex Cover Solutions· AI
- Small Language Models Show Significant Vulnerability to Jailbreak Attacks, Study Finds· AI Security
- New Architecture Helps Language Models Better Process Images and Audio by Balancing Semantic and Perceptual Detail· AI
- When LLMs Judge Other LLMs: Why Expert Verification Still Matters· AI
- New technique improves AI's ability to spot tiny structures in ultrasound images· AI
- New Framework Merges Language Models and Graph Networks for Better Text-Based Graph Analysis· AI
- New framework reveals three tiers of computational ability in language models and human cognition· AI
- New dimensionality reduction method balances local detail with global structure· Machine Learning
- Researchers predict rescue team success by analyzing what members say to each other· AI
- New Benchmark Tackles Graph Neural Networks in Heterophilic Text-Rich Graphs· AI
- New framework narrows the gap in AI alignment by refining value functions iteratively· AI
- New multimodal AI system predicts drug combination safety from lab data before clinical trials· AI
- New framework relaxes excitation requirements for distributed parameter learning· Machine Learning
- Study Compares In-Context Learning and Fine-Tuning for Smaller Multilingual Models· AI
- Theory explains double descent in neural networks using path regularization· AI
- LLM-Advisor boosts robot path planning by finding cheaper routes across varied terrain· Agentic AI
- New Coaxial Drone Design Promises Better Control for Close-Range Tasks· Robotics
- Large language models show human-like grasp of rules, study finds· AI
- AI medical diagnosis tools show strong consistency but vulnerable to irrelevant information· AI
- New Watermarking Technique Protects Speaker Verification Datasets from Unauthorized Use· AI Security
- AI models can gauge urban safety from images but embed hidden demographic biases· AI
- New AI method tracks human pose using everyday music instead of specialized sensors· AI
- New Lightweight Neural Network Approach Brings Real-Time Speech Enhancement to Low-Power Devices· AI
- New Method Makes AI Training More Reliable Across Different Tasks· AI
- Physicists Map Semantic Compression Using Statistical Mechanics and Spin Glass Theory· AI