Severe weather events, such as heavy rainfall, are becoming increasingly commonare on the rise worldwide. Reliable assessments of these events can save lives and protect property. Researchers at the Karlsruhe Institute of Technology (KIT) have presented developed a new method that uses artificial intelligence (AI) to convert low-resolution global weather data into high-resolution precipitation maps. The method is fast, efficient, and independent of location. Their findings have been published in npj Climate and Atmospheric Science: https://doi.org/10.1038/s41612-025-01103-y 

While automatized vocabulary knowledge has been shown to strongly predict general listening ability, its role in speech production remains unknown. Now, researchers have compared conscious recognition of simple word meanings (declarative knowledge) with instant, automatic access to contextualized word meanings (automatized knowledge) in Japanese students learning English as a second language. The findings revealed that automatized vocabulary knowledge better predicts fluency, and practicing vocabulary in real-life contexts may help learners speak more smoothly and confidently.

CZI announced its latest AI model, GREmLN, aimed at helping researchers better understand how cells behave by focusing on the key networks that control cell behavior.

Researchers from Seoul National University (Prof. Seung-Kyun Kang), the Korea Institute of Science and Technology, KIST (Dr. Hyojin Lee), Kwangwoon University (Prof. Jeonghyun Kim), and Gyeongsang National University (Prof. Seongchan Kim) have developed a wireless implantable drug delivery system that enables anticancer drugs to penetrate deep into solid tumors—without harming surrounding healthy tissue. The multidisciplinary team, led by experts in materials science, bioelectronics, and pharmaceutical engineering, offers a new strategy to enhance the efficacy of chemotherapy while minimizing side effects.

ChemELLM, a 70-billion-parameter LLM tailored for chemical engineering, outperforms leading LLMs (e.g., Deepseek-R1) on ChemEBench across 101 tasks, trained on ChemEData’s 19 billion pretraining and 1 billion fine-tuning tokens, accelerating lab-to-fab innovation.