The widespread adoption of large language models as writing tools has created a new problem for crowdsourcing: workers may quietly hand off their tasks to LLMs, producing data that no longer reflects genuine human judgment. A new arXiv paper, updated as a replacement, frames this as a direct challenge to the validity of crowdsourced datasets.
To gauge how the field is coping, the authors conducted a survey of 155 participants. The abstract describes the survey's purpose—understanding how the community addresses LLM outsourcing—but does not disclose findings or conclusions. The paper is positioned as a community survey on human data collection in the LLM era.
Because the abstract stops there, no specific results or recommendations can be reported from this source alone. The paper's significance lies in documenting a growing concern and the field's early attempts to respond, though the effectiveness of those responses remains unclear from the available text.