Post #24660

Important new paper using the staggered geographic rollout of LLM tools within Change.org to demonstrate that they didn't improve (and overall slightly harmed outcomes), while making the petitions themselves more homogeneous in language use. This has now been documented in other studies as well, adding further weight to the claim that these tools have limited or negative value in a vast majority of applications https://arxiv.org/abs/2511.13949