Position: Challenges and Future Directions of Data-Centric AI Alignment

Authors: Min-Hsuan Yeh, Jeffrey Wang, Xuefeng Du, Seongheon Park, Leitian Tao, Shawn Im, Yixuan Li

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this position paper, we highlight key challenges associated with both human-based and AI-based feedback within the data-centric alignment framework. Through qualitative analysis, we identify multiple sources of unreliability in human feedback... We conduct an in-depth qualitative study using a subset of data from the popular Anthropic-HH dataset (Bai et al., 2022a)...
Researcher Affiliation Academia 1Department of Computer Science, University of Wisconsin Madison, WI, USA. Correspondence to: Min-Hsuan Yeh <EMAIL>, Yixuan Li <EMAIL>.
Pseudocode No The paper describes a qualitative analysis and proposes future research directions, but it does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository for the methodology described.
Open Datasets Yes We conduct an in-depth qualitative study using a subset of data from the popular Anthropic-HH dataset (Bai et al., 2022a), where each question is paired with two responses: chosen and rejected by humans.
Dataset Splits No We randomly sample 80 data points from both harmless split and helpful split of Anthropic-HH dataset, and hire three annotators to re-label these 160 samples and record their thoughts and criteria during the annotation process. The paper describes how a subset was sampled for their qualitative analysis, but it does not provide specific training/test/validation splits for machine learning experiments.
Hardware Specification No The paper describes a qualitative analysis and proposes future research directions; it does not report on computational experiments that would require specific hardware specifications.
Software Dependencies No The paper discusses concepts and qualitative analysis, not the implementation of a software system with specific dependencies and version numbers.
Experiment Setup No The paper describes a qualitative study and its annotation setup in Appendix A, but it does not include details such as hyperparameters, optimizer settings, or system-level training configurations, as it does not involve training models.