January 2026 — page 1 of 2
- Hybrid Approach Combines LLM Ghost Code with Symbolic Execution to Handle Complex Program Analysis· AI
- Guided AI tutoring with citations outperforms direct material access in academic performance· AI
- Researchers Use AI-Guided Fuzzing to Catch Bugs in Zero-Knowledge Circuits· Cryptography
- How Deep Networks Learn to Parse Language Rules from Statistical Patterns· AI
- New Reward Model Framework Trains Multimodal AI Across Text, Video, and Audio· AI
- Smaller models can evaluate better than they generate, new research finds· AI
- Simple grep-style code search outperforms complex AI retrieval for repository completion· AI
- LLMs show "holographic" ability to pinpoint key words early, enabling faster short-text generation· AI
- New Benchmark Tackles Machine Translation Challenges in Chinese Social Media Slang· AI
- LLMs and AI-Powered Graphs Show Promise for Automating System Design Analysis· AI
- New Method Targets Efficient LLM Reasoning Training by Sampling Hard Problems First· Agentic AI
- Semantic consistency beats token confidence for trustworthy AI scientific answers· AI
- New Framework Predicts Equipment Lifespan With Minimal Degradation Data· AI
- LLM agents can slash security testing false alarms, but results vary wildly by model and vulnerability type· AI
- New Encryption Method Shields Sensitive Data Sent to Cloud AI Services· AI
- New Framework Teaches AI Models to Speak Nine Different English Dialects Accurately· AI
- Federated LLMs with privacy-preserving adapters could enable medical AI across hospitals· AI
- New Method Tackles LLM Training Collapse by Strategically Mixing Past and Present Reasoning· AI
- New Benchmark Exposes Fragility in Multilingual Math Reasoning Across Language Models· AI
- Researchers Use Physics-Inspired Optimization to Shrink Large Language Models by Removing Key Blocks· AI
- New Dataset Lets Models Learn From User Prompts Instead of Raw Text· AI
- New Framework Reveals How RAG Systems Fail to Align Retriever and Generator Components· AI
- How a Language Model's First Language Interferes with Learning a Second One· AI
- New benchmark reveals timing gap in AI agent safety monitoring—pattern-based guardrails fail at real-time detection· AI Safety
- Small Language Models Detect Multiple Issues in Messy Code Commits Better Than Larger Rivals· AI
- New framework evaluates how well AI models can generate business process diagrams· AI
- Bifocal Attention Framework Aims to Unlock Recursive Reasoning in Language Models· AI
- New Method Cuts Visual Document Storage Needs by Up to 155x Using AI Embeddings· AI
- New method adapts speech recognition AI using only text data· AI
- Language-Specific Token Pruning Transforms Code Summarization Efficiency· AI
- Small LLM Outperforms GPT-4o Using Real-Time Policy Evolution During Reasoning· Agentic AI
- How to Design Rewards for AI Systems That Write Code· AI
- Autonomous AI Agent Finds Hidden Security Flaws Before They Mature· Agentic AI
- ReportLogic: A New Benchmark for Evaluating Logical Quality in AI-Generated Research Reports· AI
- Researchers Propose Privacy-Conscious Framework for Ads in AI Chatbot Responses· AI
- LLM-based security tools struggle at scale, study finds—despite high recall on simple code· AI
- AI agents now generate bug fixes and their test cases together, cutting engineering overhead· Agentic AI
- New mutation testing framework targets Python's dynamic typing quirks· Software Engineering
- Iterative Retrieval Outpaces Perfect Context in Scientific Question Answering· AI
- New technique adapts speech AI to underserved languages without retraining· AI
- Researchers prove Boolean query evaluation over search indices is computationally hard—and offer a fix· Information Retrieval
- Self-Teaching AI: How Models Learn to Generate Their Own Training Problems· AI
- New sampling method beats greedy decoding for extracting information from documents· AI
- Language families could unlock more efficient multilingual speech recognition systems· AI
- Neural oscillations offer new window into how bilingual speakers mix languages· Neuroscience & AI
- New Benchmark Exposes How AI Struggles With Internet Memes and Cultural Context· AI
- How speech and text share computation differently across AI models· AI
- Selective Gradients Outperform Full Projections in Explainable AI for Large Language Models· AI
- New Method Helps Large Language Models Better Identify Specific Text Spans· AI
- New Benchmark Exposes How AI Vision Models Struggle With Geometric Puzzle Assembly· AI
- New Technique Narrows Language Performance Gap in Multilingual AI Models· AI
- New White-Box Method Reveals Hidden Bias in Large Language Models That Black-Box Tests Miss· AI
- New AI Framework Advances Medical Report Generation Using Reinforcement Learning· AI
- Foundation models show promise detecting hidden compilation errors in configurable C code· AI
- LLM Framework Reveals Hardware Security Flaws by Crafting Trojans That Evade Detection· AI Security
- New guardrail model offers transparent, explainable safety checks for AI language models· Agentic AI
- DeepMind's D4RT: Unified AI Model Reconstructs Dynamic Scenes 300x Faster· AI
- New activation-based filter helps AI agents protect sensitive data without slowing down responses· AI Safety
- New AI Method Generates Realistic Test Cases for Deep Learning Systems Without Labeled Data· AI
- New Python vulnerability detector claims near-perfect accuracy with 100x speed advantage· Security
- New Framework Helps LLMs Extract Knowledge Without Manual Prompts· AI
- Researchers Map Common Bugs in LLM Agent Systems Using Automated Detection· Agentic AI
- New Metric Reveals Why AI Systems Nail Facts but Fail at Logical Reasoning· AI
- New technique improves speech AI's ability to recognize domain-specific terms and names· AI
- Physics model controls consensus and dissent on online discussion platforms· AI
- Researchers Propose System to Group Similar Counterexamples in Hybrid Systems Testing· Software Engineering
- New Framework Prevents AI Agents From Losing Their Personalities in Group Settings· AI
- New Training Framework Enables Open-Source LLMs to Generate Bug Reproduction Tests· AI
- Language Models Can Improve Accuracy Without Human Feedback via Coherence Optimization· Artificial Intelligence
- Researchers Uncover LLM Vulnerability: Drunk Language Inducement Bypasses Safety Guardrails· AI
- New Benchmark Reveals How Cultural Background Shapes AI's Emotional Understanding· AI
- New Method Enables AI to Synthesize Loud Speech Without Training Data· AI
- New framework helps AI models respect diverse human values without retraining· AI
- Study challenges ArXiv's review paper ban, finds more AI-generated content in non-review papers· AI
- OI-Bench: New Benchmark Tests How Easily LLMs Fall for Misleading Directives· AI
- New AI Model Aims to Understand Memes Across Languages and Context· AI
- Researchers Discover Critical Security Flaw in Android AI Agents That Allows App Hijacking· AI Security
- Research Agents Struggle to Find and Organize Essential Papers, Study Shows· AI
- New Dataset and Training Strategy Boost Arabic Speech AI Performance· AI
- Study Reveals What Training Approach Works Best for Code Verifier AI· AI
- LLMs shift their political interpretations based on prompt language, not just content· AI
- Study Reveals Hidden Memorization Shortcut in LLMs Trained with Flawed Rewards· AI
- New Method Lets AI Systems Adapt on the Fly to Specialized Domains· AI
- How Linear Relations Drive Language Model Hallucinations· AI
- New benchmark tests how well AI agents plan complex space missions· Agentic AI
- New Framework Enables AI to Handle Complex Multi-Character Roleplay with Dynamic Scenes· AI
- New Framework Uses AI to Pinpoint Hate Speech in Videos With Timestamps and Target IDs· AI
- New Framework Automatically Evaluates AI Research Agents Using Live Web Data· AI
- New RL Framework Overcomes LLM Translation Length Problem for Time-Sensitive Media· AI
- New metric exposes hidden weakness in neural network representations undetected by standard measures· Machine Learning
- New approach helps AI understand tonal languages with minimal training data· AI
- Study of 20,000 Real Conversations Shows Purpose-Built Mental Health AI Outperforms Leading General Models at Crisis Detection· AI Safety
- Newer LLMs Simplify Test Generation, Making Complex Techniques Obsolete· AI
- How language models handle conflicting knowledge stored internally· AI
- Benchmark for evaluating AI slide generation tackles cross-system comparability challenge· AI
- New theoretical framework reveals fundamental limits of safe language generation· AI
- SwiftMem brings fast memory recall to AI agents with smart query indexing· Agentic AI
- New Framework Helps LLM Judges Score Text Consistently Against Human Standards· AI
- Study reveals large language models struggle with reconstructing scrambled text structure· AI
- New approach lets language models handle conflicting user preferences through multiple-choice alignment· AI