A dataset comparing the performance of a publicly available generative artificial intelligence chatbot and its customised version in providing first aid guidance for seizures
Description
The applicability of generative artificial intelligence chatbots as first aid consultants is a topical issue. This dataset contains the results of an analysis comparing the quality of seizure first aid recommendations generated by the publicly available chatbot ChatGPT (GPT-4o model) with those generated by its customised version. The dataset consists of three files. The first file (customisation rules.txt) contains customised text instructions for the chatbot, including definitions of key terms and roles, communication and dialogue style guidelines, a catalogue and description of knowledge base documents, operational recommendations for applying knowledge base documents in dialogue, prohibited actions, barrier mitigation strategies, chatbot phrasing examples, and conversation closure instructions. The second file (instructions.txt) contains four sets of mandatory questions and instructional wordings corresponding to the following emergency scenarios: scenario I – an unconscious victim with ongoing seizures; scenario II – a victim in the postictal period, unconscious, not breathing; scenario III – a victim in the postictal period, unconscious, breathing normally; scenario IV – a victim in the postictal period, conscious. The third file (evaluation results.xlsx) contains the results of a comparative analysis of the effectiveness of the publicly available chatbot (Baseline_# sheets) and its customised version (Custom_# sheets) according to checklists corresponding to the dialogue scenarios.