A small transformer model trained on a 5090 achieved 44% on ARC-AGI-1 and 7% on ARC-2, outperforming many LLMs.
A small trial showed CRISPR-based treatment targeting liver proteins lowered cholesterol levels for at least a year.
New features and improvements in the updated Claude models. System card available for review.
Pentagon provides 3 million personnel with access to AI tools tailored for military use.
Multidimensional MRI detected microstructural heterogeneity in brain tissue linked to amyloid pathology, offering new insights into Alzheimer's disease.
Longevity medicine is expanding, but standards and clinical practices are still developing. Tools like epigenetic clocks and microbiome panels are now mainstream.
High-resolution imaging revealed detailed dynamics of serotonergic axon development in real time.
The method can help improve disease diagnostics and cell therapy production by analyzing immune cell metabolism.
Median survival for patients with midline glioma was 20.5 months after diagnosis with lymphodepletion.
A new imaging system combines photoacoustic and ultrasound to provide high-resolution whole-body scans with molecular details.
Lilly acquired Merida Biosciences for up to $2.9B to develop drugs that selectively degrade pathogenic autoantibodies.
ImmunoBrain’s checkpoint immunotherapy, which strengthens the immune system, has a clean safety profile. Published in Nature Medicine.
Alnylam continues research on silencing drugs for ATTR-CM despite the failure of AstraZeneca's phase 3 trial.
AIR, an AI security startup, raised $50M to build a platform that vets AI agent skills, plugins, and add-ons. The platform helps companies monitor and secure their AI toolchains.
CAR-T trials for autoimmune diseases paused due to immune-related adverse events.
Humanoid robots learn to traverse sparse 3D structures using lidar-based perception and reinforcement learning.
Researchers used gold nanoconstructs to temporarily open the blood-brain barrier under laser light, improving drug delivery.
BLA models use language to condition neural signal interpretation for robotic action generation.
Nature article discusses accountability for errors made by AI in scientific research.
LLMs generate correlated name pairs that appear across multiple documents, such as Elena Vasquez and Marcus Chen.