Paper Number
2551
Paper Type
Completed
Description
Conversational generative language models (GLMs) like ChatGPT are being rapidly adopted. Previous research on non-conversational GLMs showed that formulating prompts is critical for receiving good outputs. However, it is unclear how conversational GLMs are used when solving complex problems that require multi-step interactions. This paper addresses this research gap based on findings from a large participant event we conducted, where ChatGPT was iteratively and in a multi-step manner used while solving a complex problem. We derived a taxonomy of prompting behavior employed for solving complex problems as well as archetypes. While the taxonomy provides common knowledge on GLMs usage based on analyzed input-prompts, the different archetypes facilitate the classification of operators according to their usage. With both we provide exploratory knowledge and a foundation for design science research endeavors, which can be referred to, enabling further research and development of prompt engineering, prompting tactics, and prompting strategies on common ground.
Recommended Citation
von Brackel-Schmidt, Constantin; Kučević, Emir; Memmert, Lucas; Tavanapour, Navid; Cvetkovic, Izabel; Bittner, Eva A. C.; and Böhmann, Tilo, "A User-centric Taxonomy for Conversational Generative Language Models" (2023). ICIS 2023 Proceedings. 1.
https://aisel.aisnet.org/icis2023/generalis/generalis/1
A User-centric Taxonomy for Conversational Generative Language Models
Conversational generative language models (GLMs) like ChatGPT are being rapidly adopted. Previous research on non-conversational GLMs showed that formulating prompts is critical for receiving good outputs. However, it is unclear how conversational GLMs are used when solving complex problems that require multi-step interactions. This paper addresses this research gap based on findings from a large participant event we conducted, where ChatGPT was iteratively and in a multi-step manner used while solving a complex problem. We derived a taxonomy of prompting behavior employed for solving complex problems as well as archetypes. While the taxonomy provides common knowledge on GLMs usage based on analyzed input-prompts, the different archetypes facilitate the classification of operators according to their usage. With both we provide exploratory knowledge and a foundation for design science research endeavors, which can be referred to, enabling further research and development of prompt engineering, prompting tactics, and prompting strategies on common ground.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
02-General