Post #4195097
2026-07-29 11:03 UTC
In "Commercial Determinants of Replication Infeasibility," Cassidy Waldrip, David Lazer and I ask if fears of corporate control of computational social science have become true.
The answer? Few papers that rely on tech firms can be replicated. In contrast, studies that rely on Wikipedia are much more replicable, and the replication barriers are often due to third-party corporate data that they included in the study.
https://journalqd.org/article/view/9661
Replies (1)
-
@natematias@social.coop 2026-07-29 11:11
In "Proprietary LLMs impede scientific transparency and reproducibility," Killian McLaughlin, Alexis Palmer, Molly Crockett, and I explain how the rush for LLMs in social science is destabilizing scientific reliability. Proprietary models lack the transparency for repeatable science, and even worse - their behavior drifts over time. Alexis has shown that research findings can completely change between two runs of the "same" model over time. Our article offers suggestions, and Killian is working on a stealth project in this space. https://www.nature.com/articles/s44159-026-00593-8