April 2026 — page 1 of 2
- Google DeepMind announces AI co-clinician research to augment physician care· AI
- EviMem Improves Long-Conversation AI Memory by Diagnosing Missing Evidence· AI
- Researchers Develop Hierarchical Fault Detection System for Transformer Models· AI
- How benchmark design choices quietly reshape AI evaluation in finance· AI
- LLMs Hide Better Understanding Than They Admit—But Still Fail at Strategy· AI
- New approach tackles reward over-optimization in AI model training· AI
- How people tell their stories reveals mental health more than word choice alone· AI
- Authorship Science Reveals What LLM Personalization Benchmarks Miss· AI
- Researchers develop framework to convert code into comparable graph structures for verification reuse· Software Engineering
- New Research Warns of "Cognitive Atrophy" in AI-Dependent Software Engineering· AI
- Emotion Embeddings for Speech Synthesis Fail Human Perception Tests· AI
- New technique compresses LLM reasoning by 8% using learned "supertokens"· AI
- New Method Automatically Designs LLM Workflows Without Manual Prompting· Agentic AI
- New Benchmark Reveals Sign Language AI Models Struggle With Non-Manual Cues· AI
- Speculative Decoding Accelerates AI Code Tasks, With Different Methods Suiting Different Scenarios· AI
- EvoSelect tackles efficient LLM adaptation with smart synthetic data filtering· AI
- New Framework Tracks How LLMs Handle Harmful Requests—Revealing Compliance, Refusal, and Dangerous Escalations· AI
- New Method Detects AI Hallucinations by Analyzing Hidden Steps in Diffusion Language Models· AI
- Google DeepMind Partners with South Korea to Accelerate AI-Driven Scientific Discovery· AI
- New Framework Helps AI Models Learn Private Company Code Libraries Through Execution Feedback· AI
- New Framework Uses AI to Audit LLM Agent Benchmarks for Hidden Flaws· AI
- New system checks whether AI-generated legal code is actually accurate· AI
- LLM judges change their verdicts when prompts are reworded, new benchmark reveals· AI
- Researchers Propose KISS Sorcar, a Lightweight AI Agent Framework for Long-Horizon Engineering Tasks· Agentic AI
- Researchers teach AI to ask the right questions about scientific figures· AI
- New technique uses business semantics to catch logic bugs that code-focused testing misses· AI
- First full-duplex Hindi dialogue system handles natural conversation patterns like interruptions and overlaps· AI
- Autonomous system automates environment setup across any programming language and repository· AI
- LLMs Show Promise but Fall Short as Standalone Goal Extractors for Requirements Engineering· AI & Software Engineering
- AI Coding Agents Burn Through 1,000x More Tokens Than Simple Tasks—and Prediction Fails· Agentic AI
- New Benchmark Tests AI Code Generation Against Real Industry Practices· AI
- Researchers propose framework for what AI labs should disclose about internal model deployments· AI Governance
- New Method Preserves Rare Expert Knowledge During Fine-Tuning of Mixture-of-Experts Models· AI
- Power-law data distributions outperform uniform sampling for teaching AI compositional reasoning· AI
- New benchmark pits AI models against each other to measure true math capabilities· AI
- New GUI Automation Framework Cuts Wasted Steps by Half Through Smart Stopping and Error Recovery· Agentic AI
- New Benchmark Reveals Text Anonymization Fails to Truly Protect Personal Data· AI
- Agentic AI coding can embed domain expertise through epistemic grounding documents· Agentic AI
- DeepMind's Decoupled DiLoCo trains large models across distant data centers with minimal bandwidth and better fault tolerance· AI
- New Benchmark Reveals LLMs Struggle When Code Requirements Are Vague· AI
- New algorithm tackles memory bottleneck in large language models by smartly compressing cached data· AI
- Patched Code Still Carries Hidden Security Risks, New Study Finds· Security
- New dataset reveals how developers really use AI coding agents—and where they fall short· AI
- ShadowPEFT Offers New Approach to Fine-Tuning Large Models with Fewer Parameters· AI
- New Testing Method Uncovers Hidden Inconsistencies in Code-Writing AI Models· AI
- New Framework Fixes Logic Bugs in AI-Generated GUI Applications· AI
- New tool uses AI and execution testing to verify which library dependencies actually matter· Software Engineering
- New Method Improves AI-Generated Bug Tests by Smarter Code Retrieval· AI
- Most Android Apps Log Sensitive Data They Don't Disclose in Privacy Policies· Privacy & Security
- New method improves diffusion language models by remasking low-confidence tokens instead of replacing them· AI
- New AI Framework Teaches Agents to Build Playable Games End-to-End· Agentic AI
- New Algorithm Boosts LLM Exploration Without Model Retraining· AI
- Logging automation varies dramatically across programming languages, new benchmark shows· AI
- New Benchmark Reveals Leading AI Models Struggle with Precise Code Debugging· AI
- How Multilingual AI Models Learn to Translate: Copy-First Strategy Revealed· AI
- Do LLMs Have Minds? Researchers Map the Location of AI Consciousness Through Persona Vectors· AI
- Smaller AI Models Outperform Metrics in Customer Support—A Study Reveals Validation Loss Alone Misleads Fine-Tuning Choices· AI
- Researchers Map Security Risks in AI Agents' Persistent Memory Systems· Agentic AI
- New hybrid training method unifies supervised learning and reinforcement learning for vision-language AI· Agentic AI
- New Framework Diagnoses Why Retrieval Fails in AI Language Models· AI
- Compact "Gene" Encoding Beats Documentation for Reusable AI Problem-Solving Experience· AI
- New Protocol Shields AI Multi-Agent Systems from Conformity-Driven False Consensus· AI
- Transformers Make Fast, Irreversible Decisions Through Task-Specific Commitment Heads· AI Research
- HoloTab Brings Computer-Use AI to Everyday Browser Tasks· AI
- New framework lets humans guide AI's fact-checking reasoning in real time· AI
- Structured formats beat generic text for creating synthetic training data, study finds· AI
- New AI Framework Tackles Multi-Page Document Questions Through Active Visual Reasoning· Agentic AI
- Study reveals commit signing rarely used authentically on GitHub despite security push· Software Engineering
- New self-training method lets multiple AI models improve each other without external feedback· AI
- Research breaks down the architecture behind AI coding agents like Claude Code· Agentic AI
- New Benchmark Reveals How Well AI Models Really Reason Through Coding Tasks· AI
- Do AI agents actually follow their instructions? New study reveals widespread plan abandonment· Agentic AI
- New Interface Lets Users Control LLM Output Through Interactive Widgets Instead of Text Alone· AI
- New Technique Helps AI Code Generators Balance Safety and Functionality· AI
- AI Models Learn to Deceive Attackers by Reasoning About Their Beliefs· AI
- Large language models offer far more stable explanations than traditional text classifiers, study finds· AI
- Which Language Models Make the Best Multilingual Teachers? New Study Breaks the Bigger-Is-Better Myth· AI
- New metric separates real reasoning from lucky answers in AI models· AI
- AI Framework Validates Suspected Bugs with Automated Tests, Finding 121 Real Defects· AI
- New AI Method Scales Long-Horizon Agent Tasks by Intelligently Merging Parallel Attempts· Agentic AI
- LLM agents achieve 94% success on automated software analysis, outpacing existing baselines· AI
- Academic review examines Nix's solutions to deployment problems—and its own shortcomings· Software Engineering
- RAG systems overlook opinion diversity, risking echo chambers and minority erasure· AI
- New AI Method Dramatically Improves Detection of Hidden Vulnerabilities in Source Code· AI
- LLM-Driven Code Analyzer Uses Execution Verification to Catch Vulnerabilities Across Languages· Agentic AI
- New Generative Approach to Value Models Boosts LLM Reinforcement Learning· AI
- Researchers use AI to reconstruct why code was changed by mining scattered documentation· AI
- New Framework Optimizes AI Agent Skills for Software Tasks While Cutting Costs· Agentic AI
- AI Incident Study Reveals Major Gaps in Post-Deployment Oversight Under Global Regulations· AI Governance
- Study reveals critical validation gaps in multi-agent AI framework bugs· Agentic AI
- Direct Audio Analysis Detects Abusive Speech in Indic Languages Without Transcription· AI
- New AI tutoring framework adapts personalized learning to student needs· Agentic AI
- New "ECM Contracts" Framework Prevents Robot Software Integration Failures Before Deployment· Agentic AI
- AI models struggle to capture moral diversity across cultures, new study finds· AI
- New Research Reveals How AI Chatbots Prioritize Profits Over User Interests· AI
- New scoring method helps developers map code across redesigned software projects· Software Engineering
- Emotional Language Weakens Math Skills in AI Models—But Neutral Rewrites Help· AI Research
- ZeroCoder co-evolves AI code and test generation without labeled training data· AI
- Researchers Expose Visual Vulnerabilities in GUI Agents Through Deceptive UI Overlays· AI Safety
- Researchers Develop Method to Predict When Retrieval Augmentation Helps Question Answering· AI