technological knowledge use in papers (A, B, and C) and patents (D, E, and F) based on following measures: diversity of work cited (A and D), mean number of self-citations (B and E), and mean age of cited work (C and F). The inset plots of A and D show changes in the share of citations to the top 1% most highly cited papers (A1 and D1) and in the semantic diversity of the top 1% most cited over time (A2 and D2). Values of both measures are computed within field and year, and are subsequently averaged across fields for plotting. Semantic diversity is based on paper and patent titles; values correspond to the ratio of the standard deviation to the mean pairwise cosine similarity (i.e., the coefficient of variation) among the titles of the 1% most cited papers and patents by field and year. To enable semantic comparisons, titles were vectorized using pretrained word embeddings. For papers, lines are shown for each WoS research area; for patents, lines are shown for each NBER technology category. In subsequent regression analyses using these measures, we find that using less diverse work, more of one’s own work, and older work is associated with less disruptive papers and patents