Dynamic Robustness Profiling: Modeling Prompt-Model Interactions Over Model Updates

Building on the findings from Chang et al. (2024, 2025), who observed that a significant fraction of prompts judged appropriate in GPT-3.5 became inappropriate in GPT-4.0 and later, this research proposes systematically modeling the evolution of prompt robustness across LLM generations. Unlike current static benchmarking practices, this approach treats prompt robustness as a dynamic property, mapping how and why prompt-model interactions shift with each new model release or fine-tuning cycle. The methodology would combine longitudinal benchmarking, prompt clustering, and causal analysis tools to surface which changes in model architecture or training data most affect robustness. This could empower developers to foresee and mitigate regressions, and provide insights into the “hidden” brittleness that emerges only after deployment. It’s a much-needed extension to the static red-teaming and snapshot evaluations prevalent in today’s literature.

References:

1. Red Teaming Large Language Models in Medicine: Real-World Insights on Model Behavior. Crystal T. Chang, Hodan Farah, Haiwen Gui, S. Rezaei, Charbel Bou-Khalil, Ye-Jean Park, Akshay Swaminathan, J. Omiye, Akaash Kolluri, Akash Chaurasia, Alejandro Lozano, Alice Heiman, A. Jia, Amit Kaushal, Angela Jia, Angelica Iacovelli, Archer Yang, Arghavan Salles, Arpita Singhal, Balasubramanian Narasimhan, Benjamin Belai, Benjamin H. Jacobson, Binglan Li, Celeste H. Poe, C. Sanghera, Chenming Zheng, Conor Messer, Damien Varid Kettud, Deven Pandya, Dhamanpreet Kaur, Diana Hla, Diba Dindoust, Dominik Moehrle, Duncan Ross, Ellaine Chou, Eric Lin, Fateme Nateghi, Haredasht, Ge Cheng, Irena Gao, Jacob Chang, J. Silberg, J. Fries, Jiapeng Xu, Joe Jamison, John S. Tamaresis, Jonathan H. Chen, Joshua Lazaro, Juan M. Banda, Julie J. Lee, K. Matthys, Kirsten R. Steffner, Lu Tian, Luca Pegolotti, Malathi Srinivasan, Maniragav Manimaran, Matthew Schwede, Minghe Zhang, Minh Nguyen, Mohsen Fathzadeh, Qian Zhao, Rika Bajra, Rohit Khurana, Ruhana Azam, Rush Bartlett, Sang T. Truong, Scott L. Fleming, Shriti Raj, Solveig Behr, Sonia Onyeka, Sri Muppidi, Tarek Bandali, Tiffany Eulalio, Wenyuan Chen, Xuanyu Zhou, Yanan Ding, Ying Cui, Yuqi Tan, Yutong Liu, Nigam H. Shah, Roxana Daneshjou (2024). medRxiv.
2. Red teaming ChatGPT in medicine to yield real-world insights on model behavior. Crystal T. Chang, Hodan Farah, Haiwen Gui, S. Rezaei, Charbel Bou-Khalil, Ye-Jean Park, Akshay Swaminathan, J. Omiye, Akaash Kolluri, Akash Chaurasia, Alejandro Lozano, Alice Heiman, A. Jia, Amit Kaushal, Angela Jia, Angelica Iacovelli, Archer Yang, Arghavan Salles, Arpita Singhal, Balasubramanian Narasimhan, Benjamin Belai, Benjamin H. Jacobson, Binglan Li, Celeste H. Poe, C. Sanghera, Chenming Zheng, Conor Messer, Damien Varid Kettud, Deven Pandya, Dhamanpreet Kaur, Diana Hla, Diba Dindoust, Dominik Moehrle, Duncan Ross, Ellaine Chou, Eric Lin, F. N. Haredasht, Ge Cheng, Irena Gao, Jacob Chang, J. Silberg, Jason A. Fries, Jiapeng Xu, Joe Jamison, John S. Tamaresis, Jonathan H. Chen, Joshua Lazaro, Juan M. Banda, Julie J. Lee, K. Matthys, Kirsten R. Steffner, Lu Tian, Luca Pegolotti, Malathi Srinivasan, Maniragav Manimaran, Matthew Schwede, Minghe Zhang, Minh Nguyen, Mohsen Fathzadeh, Qian Zhao, Rika Bajra, Rohit Khurana, Ruhana Azam, Rush Bartlett, Sang T. Truong, Scott L. Fleming, Shriti Raj, Solveig Behr, Sonia Onyeka, Sri Muppidi, Tarek Bandali, Tiffany Y. Eulalio, Wenyuan Chen, Xuanyu Zhou, Yanan Ding, Ying Cui, Yuqi Tan, Yutong Liu, Nigam H. Shah, Roxana Daneshjou (2025). npj Digit. Medicine.