Researchers pioneer new technique to stop LLMs from giving users unsafe responses
Peer-Reviewed Publication
Updates every hour. Last Updated: 2-Apr-2026 07:15 ET (2-Apr-2026 11:15 GMT/UTC)
Researchers have identified key components in large language models (LLMs) that play a critical role in ensuring these AI systems provide safe responses to user queries. The researchers used these insights to develop and demonstrate AI training techniques that improve LLM safety while minimizing the “alignment tax,” meaning the AI becomes safer without significantly affecting performance.
A new study reviews how machine learning (ML) is being used to help communities recover critical infrastructure after natural hazards such as earthquakes, floods, and hurricanes. The research synthesizes global studies and shows that ML can support recovery by characterizing recovery trends, predicting recovery times, and optimizing recovery schedules. The authors also identify key challenges, such as limited data availability, and outline future directions for building more resilient infrastructure systems using ML.
Cryopreservation is not a new technology, but there is still much to explore and perfect in the field. Current methods use slow freezing, a method that is conducive to ice formation, cell dehydration and an increase in cryoprotective agents (CPAs). These are not ideal circumstances for achieving immaculately cryopreserved cells. Researchers from the University of Tokyo use vitrification, a process that transforms a substance into a noncrystalline solid by rapid cooling. This cooling yields favorable outcomes in biological samples, even those that are typically difficult to freeze and thaw successfully. Despite challenges within this method, the future of regenerative medicine research may be greatly, and positively, impacted by the use of vitrification for cell cryopreservation.
A newly developed 2.4 GHz Wi-Fi receiver from Science Tokyo can survive radiation levels found inside nuclear reactors. With a radiation tolerance of up to 500 kGy, the chip allows robots used in nuclear plant decommissioning to be controlled wirelessly. Such receivers reduce the need for wired connections and can improve worker protection during decommissioning and cleanup operations at contaminated sites such as the Fukushima Daiichi Nuclear Power Plant.
Researchers have designed a nitrogen-rich porous aromatic framework material and investigated its electrochemical performance as the cathode material for sodium organic batteries. The aromatic framework material synthesized by introducing the redox-active hexaazatrinaphthylene (HATN) motif has a high redox potential and multi-ion storage capacity, and can still maintain a high capacity and excellent stability within the temperature range of -20 °C to 50 °C.