April 2025
- Structured AI literacy training shifts how CS students approach generative AI responsibly· Education & AI
- Survey maps emerging fairness challenges in graph learning augmented by machine learning· AI
- Backdoor Attack Defeats AI Image Model Safety Techniques, Study Finds· AI Security
- First study maps LLM supply chain vulnerabilities and structural risks· AI
- New LLM Framework Generates Complex Hardware Designs with 7B-Parameter Models· AI
- Reflexion: An AI prototype for guided emotional self-exploration· AI
- AI agent kAgent learns to fix Linux kernel crashes like human developers· AI, Software Engineering
- When Should Organizations Let AI Lead? Study Challenges "AI-First" Thinking· AI
- TreeHop: A faster way to answer complex questions without relying on LLMs· AI
- New compression-based technique eliminates repetitive text in AI language models· AI
- AI models crack computational number theory with near-perfect accuracy on algorithmic tasks· AI
- LLMs Show Promise for Mining Lunar Resource Data from Scientific Literature· AI
- Framework Maps Three AI Roles for Human-Machine Collaboration in Autonomous Vehicles· AI
- New Framework Lets Users Fine-Tune AI Safety Without Retraining Models· AI
- Graph Neural Networks Boost Kinship Verification Accuracy from Facial Images· AI
- AI-Powered Actor-Critic Model Optimizes Network Function Splitting for 6G· Agentic AI
- Symmetric Matrices Unlock New Pathways for Brain Imaging and AI· AI
- New Graph-Based AI Model Speeds Up Ontology Checking by 20x· AI
- Reinforcement Learning Approach Improves NLI Models' Ability to Handle Adversarial Data· AI
- Pleias-RAG: Compact Models Prove Small Reasoners Can Ground Answers in Sources· AI
- New benchmark systematically compares methods to measure and reduce uncertainty in large language models· AI
- New robot planning framework combines black-box skills for complex multi-step tasks· Robotics
- New Multi-Agent AI Method Speeds Up Feature Selection Using Hierarchical Reinforcement Learning· Agentic AI
- Framework reveals how psychological safety shapes open-source collaboration patterns· Open Source Software
- New EM-based framework simplifies reinforcement learning for reasoning AI models· AI
- New LLM-based tool mines REST API test constraints from specs rather than code execution· AI
- Study reveals how LLMs shift their political stances depending on the topic· AI
- New LLM Approach Automates REST API Documentation, Catching Gaps Developers Miss· AI
- Researchers develop one-shot learning system to detect calls from critically endangered pigeons· AI
- Malware Now Dominates Open Source Vulnerability Records, Study Finds· Security
- Withdrawn Proposal Outlines Framework for Testing LLMs on Mandarin Grammar· AI
- AI Model Fixes Broken Software Requirements Links in Automotive Systems· AI
- Study reveals how AlphaZero internalizes search strategy into learned behavior· Machine Learning
- New Framework Merges AI with Human Reasoning to Amplify Decision-Making· AI
- New mathematical framework reveals how spiking neural networks make decisions· AI
- Survey Maps the Frontier of Efficient Continual Learning for Large AI Models· AI
- Philosopher argues advanced AI systems possess genuine understanding and cognitive abilities· AI
- New Benchmark Exposes LLM Weaknesses in Web3 Security and Financial Risk· AI
- Researchers develop lightweight AI system to improve childhood tumor diagnosis from pathology images· AI
- Study reveals hard ceiling for LLM reasoning: why adding compute alone won't fix non-reasoning models· AI
- New Tool Catches Hidden Flaws in GraphQL APIs That Standard Security Scanners Miss· Security
- New technique helps AI learn safety constraints from rough trajectory feedback· AI
- Plain Transformers Emerge as Effective Graph Neural Networks Without Added Complexity· AI
- New benchmark reveals large language model agents struggle with Earth observation tasks· Agentic AI
- New system enables real-time speech translation with lightweight, one-way neural design· AI
- New AI System Uses Bus Cameras to Track Passenger Flow Without Tracking Individuals· AI
- New QA-Based Method Evaluates Machine Translation Without Knowing the Target Language· AI
- Multi-agent systems unlock better test-time reasoning for AI models, new research shows· Agentic AI
- New clinical AI model rivals GPT-4 in disease diagnosis across languages· AI
- AI University: Tailored Learning Assistant Outperforms Generic Models in Engineering Courses· Agentic AI
- Visual Salamandra: A 7B Multimodal Model Bridging Vision and Language with European Language Support· AI
- Sparse Autoencoders Unlock More Efficient Unlearning in Large Language Models· AI Safety
- Hybrid AI-Physics Approach Tackles Radar Elevation Errors Over Greenland Ice· AI
- How Scientific Workflows Harness Generative AI Despite Its Hallucinations· AI
- Researchers release MedHal dataset to tackle AI hallucinations in medical AI· AI
- LLMs Are Reshaping Scientific Writing With Denser Vocabulary and Simpler Sentences· AI
- New method evaluates AI chatbots by searching the web for facts instead of comparing to preset answers· AI
- New method maps causal chains in stories better than GPT-4o or Claude· AI
- Popular Commonsense Reasoning Benchmarks Hide Serious Validity Flaws· AI
- Two-Dimensional Curriculum Learning Improves Language Model Alignment Efficiency· AI
- Study finds large language models may develop computational structures of psychopathology· AI
- Tropical Neural Networks Cut Computational Cost Without Sacrificing Accuracy· AI
- New Framework Uses Machine Unlearning to Cut Costs of Training Aligned AI Models· AI
- Mode Connectivity Reveals Hidden Structure in Machine Unlearning· Machine Learning
- New AI framework learns to track and predict the behaviour of satellites and space debris· AI
- SkillFlow enables AI agents to find and use the right coding skills from massive community libraries· Agentic AI
- Unified AI Framework Tackles Over 100 Low-Level Vision Tasks with Text and Visual Prompts· AI
- Study reveals tensions between freelance writers and AI tools over equity, job security· AI
- Causal Attribution Can Catch LLMs Gaming Explanations for Rewards· AI
- New approach harnesses LLMs for better document retrieval using query likelihood modeling· AI
- New Framework Helps AI Models Adapt to Shifting Domains While Remembering Past Knowledge· AI
- AI models learn by generating their own knowledge through environment interaction· AI
- Research roadmap charts path for AI foundation models in cyber-physical systems engineering· AI
- Researchers Develop Predictive Formulas to Guide Language Model Compression· AI
- Hyperflux: A New Framework for Understanding and Controlling Neural Network Pruning· AI
- AI's Role in Software Architecture Still Poorly Understood, New Review Reveals· AI
- New framework improves how AI systems weigh conflicting evidence to make better decisions· AI
- New framework enables smaller LLMs to match GPT-4 performance on database query tasks· AI
- New Framework Harnesses AI to Score Student Assessments Across Multiple Subjects· AI
- ZoomV: Smart Selective Video Watching for Understanding Long Videos Faster· AI
- FlockMTL extends DuckDB with native language models and retrieval-augmented generation· AI