June 2025 — page 2 of 2
- Researchers develop automated red-teaming framework to uncover diverse attack strategies against LLMs· AI Safety
- HauntAttack: New vulnerability in reasoning models exploits step-by-step logic· AI
- LLM-Powered Text Analysis Shows Promise for Spotting Depression Signals Online· AI
- Study identifies simplest reward design for learning from expert examples· Machine Learning
- Researchers simulate one billion AI agents to model global human behavior· AI
- New Method Hardens Deep Learning Against Adversarial Attacks Without Retraining· AI
- New Dataset Targets Gaps in AI Reasoning Over Molecular and Medical Information· AI
- OpenAg framework aims to bring tailored AI decision-making to smallholder farmers· AI
- Automated Tool Generates Security Exploits for npm Vulnerabilities at Scale· AI
- How Small Changes in Testing Methods Can Inflate LLM Performance Claims· AI
- Framework models full patient hospital stays from electronic health records· AI in Healthcare
- NTNU's Hybrid Approach Combines Audio and Language Models to Score Spoken English Proficiency· AI
- New method enables robots to learn complex tasks in under an hour of real-world training· Robotics
- New Video Reasoning Benchmark Exposes Gaps in AI Models' Ability to Understand Complex Visual Content· AI
- New benchmark tests multimodal AI on humanities and social sciences reasoning· AI
- AI matches doctors at spotting macular degeneration but struggles to predict patient outcomes· AI & Healthcare
- Open-Weight LLMs Outperform Rule-Based Systems at Extracting Disease Labels from CT Scans· AI
- EU AI Act's research carve-outs risk upending how scientists publish findings· AI Policy
- New Training Method Boosts RAG Systems' Ability to Use Retrieved Information Reliably· AI
- Sound on-device: How Arm CPUs power real-time AI audio generation· AI
- Hybrid Planning Framework Tackles Safety and Interpretability in Dual-Arm Robot Control· Robotics
- Study challenges assumption that training data leaks always inflate code AI model performance· AI
- Holo1: Open-Source Vision Models for GUI Automation Achieve 76.2% Accuracy· Agentic AI
- WebChoreArena Pushes Web Agent Benchmarking Beyond Simple Tasks· AI
- Study proposes human-centered evaluation framework for AI models beyond standard benchmarks· AI Evaluation
- LLM-powered memory system makes AI agents far more data-efficient learners· Agentic AI
- NLP Techniques Reshape Understanding of Biological Sequences Across Genomics and Proteomics· AI
- How Human Beliefs About AI Shape Preference Labels in Reinforcement Learning· AI
- New AI Framework Enables Robots to Apply Skills to Novel Tasks Without Retraining· AI
- Efficient Deep Learning on Massive Cell Datasets: scDataset Balances Speed and Quality· AI
- New method uses statistical divergence to spot false claims in AI language models· AI
- Researchers argue smaller language models are better suited for AI agents than large models· Agentic AI
- New Open-License Dataset Offers Two Trillion Tokens for Ethical LLM Training· AI
- New Diffusion Language Models Bridge the Gap Between Speed and Quality with Causal Attention· Agentic AI
- Researchers Cut FastText Memory Use by 80% With Trie-Based Compression· AI
- New AI Framework Learns Geography Through Human Movement Patterns· AI
- New AI Framework Blends Concepts Into Videos While Keeping Motion Intact· AI