Chirag SHAH - 27 mars 2025 à 10h30 au PK-5115 (201, ave President-Kennedy, 5e étage)
TITRE : Optimizing LLM Prompts for Scientific Use
RÉSUMÉ
As large language models (LLMs) increasingly permeate scientific research, their use for generating or analyzing data often relies on ad-hoc decisions, raising concerns about transparency, objectivity, and rigor. This talk introduces a methodology inspired by qualitative codebook construction to systematize prompt engineering. By integrating humans in the loop and a multi-phase verification process, this approach enhances replicability and trustworthiness in using LLMs for data analysis. Practical examples will illustrate how rigorous labeling, deliberation, and documentation can reduce subjectivity and ensure more robust and generalizable research outcomes.
BIOGRAPHIE
Chirag Shah is Professor in the Information School at the University of Washington, where he conducts research at the intersection of information retrieval, human-computer interaction, and artificial intelligence. His recent work focuses on prompt engineering for optimizing interactions with large language models, exploring both theoretical underpinnings and practical applications. Dr. Shah has authored numerous publications and is actively involved in advancing the understanding of how humans and AI systems can collaborate more effectively.
RÉFÉRENCES:
Sahoo, Pranab, et al. (2024) "A systematic survey of prompt engineering in large language models: Techniques and applications." arXiv preprint arXiv:2402.07927 (2024).
Shah, C. (2024). From Prompt Engineering to Prompt Science With Human in the Loop. arXiv preprint arXiv:2401.04122.
White, R. W., & Shah, C. (2025). Information Access in the Era of Generative AI. Springer