LLMs in the Generation of Seismic Alert Communiqués

Peru is in the Ring of Fire, a zone of high seismic activity. Currently, alerts generated by technical-scientific entities are often bland and lack precise geographic context, resulting in alerts of limited usefulness for informing the public. In this paper, we present a conceptual model and architecture to explore the potential of Large Language Models (LLMs) to produce various forms of seismic warnings tailored to the particularities that would be required for different geographic areas of a given locality in Peru. The proposal was evaluated in a controlled environment with the participation of 47 users with diverse ethnographic characteristics. The context of the study was explained to them, and they were provided with a questionnaire designed to assess the ease of understanding, usefulness and quality of the content of the alert communications generated by an LLM. The results show that, according to the indicators assessed, seismic warnings generated by an LLM are 76% easy to understand, 81% useful and 71% acceptable quality.