February 2024
- New Syntactic Backdoor Attack Threatens Pre-trained Language Models Across Tasks· AI Security
- Withdrawn arXiv Study Examined Language Model Performance on Low-Resource Aminoacian· AI
- New model reconciles biological plausibility with learning complex neural patterns· AI
- New Method Detects Poisoned Updates in Federated Learning Without Needing Validation Data· AI Security
- Researchers prove two-tower recommendation systems converge to optimal performance· AI
- Study Tests Whether AI Language Models Can Realistically Mimic Human Personalities· AI
- Novel Memoroid Framework Boosts Recurrent Reinforcement Learning Efficiency· AI
- New Datasets and GPT-4 Boost Grammar Correction for Esperanto· AI
- Researchers develop methods to train AI safely with corrupted human feedback· AI
- Researchers Propose Formal Framework for Embedding Human Values into Multi-Agent AI Systems· AI
- New Statistical Method Evaluates Generative AI Quality Without Training Extra Models· AI
- Researchers Demonstrate Attacks on Multi-Agent AI Systems Using Only Partial Information· AI Security
- Researchers solve decidability puzzle in temporal reasoning for AI systems· Artificial Intelligence
- New AI System Teaches LLMs to Understand and Reason About 3D Sound Environments· AI
- AI models extract sex-based blood pressure data from medical literature· AI
- Researchers Test Multi-Agent AI for Real-World Code Generation, Find Promise but Practical Hurdles Remain· Agentic AI
- New Method Uses Human Psychology to Align AI Models More Effectively· AI
- New RAG Framework Tackles Retrieval Challenges in Multi-Document Question-Answering· AI
- New Survey Maps the Landscape of Tractable Probabilistic Models· Machine Learning