December 2025
- New method improves how language models use tools by learning interpretable rules from failures· AI
- Smart Models Learn When to Think Harder: Training-Free Gating Strategy Boosts Code Generation· AI
- New model teaches AI to understand human emotions as a driver of future actions· AI
- How software engineers are actually using AI—and what's holding them back· AI
- Theoretical Bounds for Private and Robust Language Model Alignment· AI
- Turkish NLP Gets a Modern Upgrade: TabiBERT Encoder Outperforms Predecessors on New Benchmark· AI
- New LLM Framework Aims to Speed Up Python Code Without Hand-Coded Rules· AI
- New Research Shows LLMs Can Be Tricked Into Building Software Exploits Without Technical Expertise· AI Security
- Fine-Tuned AI Models for Code Patching May Be Weaker Than They Appear· AI
- Study Develops Framework to Assess Privacy Risk of Health Data in Online Medical Conversations· AI
- Survey Maps LLM-Based Agents for Automating Software Bug Fixes and Issue Resolution· Agentic AI
- New AI Model Learns Moral Judgment by Understanding Context, Outperforming Standard LLMs· AI Alignment
- Cutting Training Sequences in Half Maintains AI Model Performance While Slashing Computational Costs· AI
- Transparent AI Essay Grader Lets Teachers Inspect and Override Scoring Decisions· AI
- Preserving Raw Conversation Outperforms Structured Memory Extraction for Long-Context LLMs· AI
- Withdrawn paper proposes geometric memory model to solve LLM context limits· AI
- Transformers encode mental states as continuous spectrum, study finds· AI
- MauBERT brings universal phonetic understanding to speech AI across 55 languages· AI
- LLM Digital Twins Show Promise but Struggle to Match Human Psychology Across Tasks· AI
- Study reveals how inference-time sampling with LLM judges balances computational gains against reward misspecification· AI
- New Tool Helps Authors Map Out AI-Generated Interactive Stories· AI
- New AI System Pinpoints Crisis Locations from Social Media Posts in Real Time· AI
- New Security-Focused Dataset Trains AI Code Generators to Build Safer Software· AI
- Vocabulary scaling patterns diverge between real video conversations and fictional dialogue· Linguistics & AI
- New Open Toolkit Tackles Fragmentation in Real-Time Speech Translation Testing· AI
- New Benchmark Tackles Evaluation of AI-Powered Research Report Generation· AI
- Researchers Introduce Canon Layers to Improve Information Flow in Language Models· AI
- Flaky quantum tests are rare but hard to catch—study of Qiskit Terra reveals hidden reliability risks· Software Engineering
- Diffusion Models Struggle With Competing Concepts in Multi-Concept Image Generation· AI
- AI Agent Learns to Write Its Own Forensics Tools for Detecting Fake Images· Agentic AI
- LLM Agents Can Leak Secrets Through Tool Combinations, New Research Shows· AI Security
- SpIDER Method Boosts Code Search for AI-Powered Bug Fixing by Exploiting Codebase Structure· Agentic AI
- New dataset and methods help AI recognize emotions expressed by sign language users· AI
- New Expert-Labeled Dataset and AI Framework Boost Music Emotion Recognition Accuracy· AI
- New System Converts Plain Language to Database Queries 27× Faster, at Penny-Per-Question Cost· AI
- New Fine-Tuning Method Cuts GPU Memory Use in Half While Matching QLoRA Accuracy· AI
- New benchmark reveals gap between visual polish and reasoning in AI image and video generators· AI
- New framework helps AI reasoning models waste fewer tokens on unnecessary steps· AI
- New benchmark measures AI's ability to turn business goals into database products· AI
- Expert Developers Keep the Reins: Study Shows Programmers Use AI Agents for Speed, Not Surrender· Agentic AI
- New Framework Measures How Well Music Editing Tools Preserve Original Context· AI
- New Method Reveals How AI Models Mishandle Sensitive Reproductive Health Advice· AI
- ATLAS: Open-Source Tool Generates Unified Code Graphs for C and C++ Without Full Compilation· Software Engineering
- New Framework Makes Knowledge Graph Reasoning Faster and More Interpretable for Sparse Data· AI
- New AI Framework Cuts Video Processing Costs by Intelligently Selecting Key Frames· AI
- AI Tools Boost Financial Analysts' Research Breadth—But Paradoxically Hurt Forecast Accuracy· AI
- New tool extends Alloy formal method to verify complex security and concurrency properties· Software Engineering
- AI Safety Checks Flake on Edge Cases—High Temperature Triggers Inconsistent Refusals· AI Safety
- Lab benchmarks don't predict real-world success for AI vulnerability detectors· AI Security
- H2Rec: Balancing Semantic and Hash IDs in Sequential Recommendation Systems· AI
- AI Chatbot System Combines Emotion Recognition and Learning Support to Boost Student Well-Being· AI
- Google DeepMind expands safety partnership with UK AI Security Institute· AI
- LLM Agent Solves Olympiad Geometry Problems with Reinforcement Learning, Beating Human Gold Medalists· Agentic AI
- Study Maps How Developers Tackle Privacy Law Compliance in Open Source· Software Engineering
- New AI Framework Automates Detection of Overlooked Findings in Abdominal CT Scans· AI
- New Framework Bridges Semantic Gap in Cross-Domain Log Anomaly Detection· AI
- ARTEMIS platform automatically tunes LLM agent configurations, boosting performance by 10-37%· Agentic AI
- New Framework Boosts Learning System Performance When Encountering Unseen Questions· AI
- Machine translation evaluation gets smarter by pinpointing errors instead of just scoring· AI
- Chemistry Foundation Model Reveals Surprising Robustness in Polymer Representation· AI
- New AI Framework Makes Error Detection in Data Transparent and Reliable· AI
- Training Language Models to Use Prolog as a Tool for Verifiable Reasoning· AI
- New framework bridges embedding geometry and human-readable topics in short-text analysis· AI
- Multi-Agent System Boosts Code Translation Accuracy to 94%· Agentic AI
- AI Framework Generates High-Performance Concrete Formulas with Strict Constraints· Agentic AI
- Smaller AI models crack security query problem at fraction of GPT-4 cost· AI
- New Method Tackles Look-Ahead Bias in Financial AI Without Retraining Models· AI
- New sparse attention method cuts LLM inference time by up to 38% without sacrificing accuracy· AI
- LLMs Generate Control Flow Graphs to Automate Test Case Creation from Use Cases· Software Engineering
- Open-source LLMs debate to auto-label sensitive mental health and safety data· AI
- New auditing framework scrutinizes how AI systems claim to understand meaning· AI
- New benchmark tackles conversational UI coding with visual feedback· AI
- GitHub's CODEOWNERS Feature Reshapes Code Review Accountability, Study Shows· Software Engineering
- New Attack Breaks Safety Defenses in Open-Source AI Models—But Researchers Have a Fix· AI Safety
- Eli Lilly's Model Gateway brings MLOps to pharmaceutical drug discovery at scale· AI
- New Benchmark Exposes Large Language Models' Struggle With Telecom Technical Tables· AI
- Researchers Build Multilingual Medical AI to Answer Questions Beyond English· AI
- Researchers Test AI Pipeline to Build Personalized Therapy Networks from Session Transcripts· AI
- Study maps human-centered design principles for AI adoption in software teams· AI
- Researchers Bridge Gap Between Continuous Sensor Data and Automated Process Control· Software Engineering
- Fast, Influence-Based Method Removes Unwanted Behaviors from Large Language Models· AI
- New framework shows how flexible behavior emerges from a single mathematical principle· AI
- Agentic AI Fuzzing System Finds Software Vulnerabilities 25x Faster Than Conventional Methods· Agentic AI
- DZ-TiDPO: A Framework for Reducing State Inertia in Long-Context Dialogue· AI
- AugServe improves augmented LLM serving efficiency with adaptive scheduling and dynamic batching· AI
- LLM Framework Automates Fuzzing Harness Generation at Scale· AI
- Belobog: New Fuzzing Tool Catches Move Smart Contract Bugs at Scale· Blockchain Security
- AI System Uncovers Hidden Structure in Prediction Markets with 63% Accuracy· Agentic AI
- Feedback loops level the playing field for LLM-based code translation systems· AI
- LLM-Generated Code Poses Major Security Risks in Real-World Applications· Agentic AI
- Pianist Transformer: Self-Supervised Learning for Expressive Piano Performance· AI
- African languages struggle in AI detection models due to sparse, code-switched online data· AI
- Researchers propose phase-based neural networks that treat language like wave phenomena· Machine Learning
- New Platform Bridges Software Ecosystem Silos to Speed Up Supply Chain Risk Assessment· Software Engineering
- New Dual Randomized Smoothing Method Breaks Trade-off Between Small and Large Adversarial Perturbations· AI