Philip S. Yu
Bridging four decades of data-mining methodology to the question of how to evaluate large language models without reinventing techniques the field already has.
Yu's published record covers most of what counts as data mining since the field had that name — stream mining, sequential pattern discovery, knowledge graphs, anomaly detection — produced from IBM Research in its long tenure and then continued at UIC, with several hundred patents along the way. His co-authorship of "A Survey on Evaluation of LLMs" reads as that body of work meeting the LLM-evaluation literature from above: many of the statistical and methodological questions that current LLM evaluation faces have been answered, partially, by techniques developed decades ago for similar problems on smaller-scale data. Reading him is the way to discover that some current "novel" evaluation methods are rediscoveries.