Extended Data Fig. 3: In-depth Analysis of Section Word Counts in Model Cards.
From: Systematic analysis of 32,111 AI model cards characterizes documentation practice in AI
(a) Comparative Assessment of Average Section Lengths in Model Cards Based on Word Count. This figure displays the average section length, measured in word count, among completed sections for all model cards, the top 1000 model cards, and the top 100 model cards. Sections such as How to Start, Training, and Limitations are substantially longer, while Citation, Evaluation, Environmental Impact, and Intended Uses are relatively shorter. Interestingly, despite its lower completion rate, the Limitations section exhibits one of the highest average word counts (161 words in the top 1000 model cards). (b-c) Disparate Community Attention Patterns Across Model Card Sections, Analysed for both the top 100 model cards (b) and all model cards (c). The Environmental Impact section demonstrates both a low completion rate and a low average word count, indicating limited community attention. In contrast, the Training section displays high completion rates and average word counts, signifying greater community engagement.
