Large language models prioritize helpfulness over accuracy in medical contexts
Peer-Reviewed Publication
Updates every hour. Last Updated: 21-Dec-2025 06:11 ET (21-Dec-2025 11:11 GMT/UTC)
Mass General Brigham-led study found that large language models (LLMs) often fail to challenge illogical medical prompts due to sycophantic behavior, posing risks for misinformation. The researchers showed that targeted fine-tuning can significantly improve model reasoning.
In a surprising discovery, MIT scientists found tiny loops in the genomes of dividing cells. Enabled by a high-resolution mapping technique, the findings overturn a long-held belief that the genome loses its 3D structure during mitosis.
Scientists at the University of Utah and Sethera Therapeutics have discovered an enzyme that can reshape therapeutic peptides—including GLP-1 drugs used for diabetes and obesity—into more stable, ring-shaped forms. New research could help accelerate drug development.