Critical Jailbreak Flaw Hits All Major LLM Providers
Security researchers uncover a universal jailbreak vulnerability that bypasses safety guardrails across GPT-4, Claude, G…
📂 Research
Latest articles in Research
Anthropic Unveils Constitutional AI 2.0 Research
Anthropic publishes new Constitutional AI 2.0 paper advancing scalable oversight methods for safer, more aligned AI syst…
AlphaFold 3 Transforms Drug Discovery for Pharma
Google DeepMind's AlphaFold 3 dramatically accelerates molecular structure prediction, reshaping how pharmaceutical comp…
RIKEN Builds Brain-Like Chip Architecture for Edge AI
Japan's RIKEN institute unveils neuromorphic computing architecture that mimics biological neural networks to run AI at …
SNU Builds Lightweight Vision Transformer for Mobile Medical AI
Seoul National University researchers develop a compact Vision Transformer that runs medical imaging diagnostics on smar…
U Tokyo Sets New Bar for Robotic Dexterity
University of Tokyo researchers achieve state-of-the-art performance on robotic manipulation benchmarks using a novel le…
IISc Bangalore Achieves Breakthrough in Neural Network Pruning
Indian Institute of Science researchers develop novel energy-efficient pruning method that cuts AI model compute costs b…
New Method Learns Diffusion Model Integrals Directly
Researchers propose learning the integral of diffusion models instead of the score function, potentially transforming sa…
Why Deep Learning Works: New Theoretical Frameworks Emerge
Researchers are making progress on one of AI's biggest mysteries — why deep neural networks generalize so well despite c…
Microsoft Cuts LLM Hallucinations by 90% With New Method
Microsoft Research unveils a novel framework that reduces large language model hallucinations by up to 90%, potentially …
Google Titans Architecture Could Replace Transformers
Google Research unveils Titans, a new neural architecture with learned memory modules that may succeed the Transformer m…
Anthropic Maps Claude's Mind With Interpretability
Anthropic researchers use mechanistic interpretability to extract millions of interpretable features from Claude, reveal…