arXiv:2507.03373v2 Announce Type: replace Abstract: Given Wikipedia's role as a trusted source of high-quality, reliable content, concerns are growing about the proliferation of low-quality machine-generated text (MGT) produced by large language models (LLMs) on its platform. Reliable detection of MGT is therefore essential. However, existing work primarily evaluates MGT detectors on generic generation tasks rather than on tasks more commonly performed by Wikipedia editors. This misalignment can lead to poor generalisability when applied in real-world Wikipedia contexts. We introduce WETBench,
The rapid proliferation and increasing sophistication of large language models (LLMs) necessitate new methods for detecting machine-generated content, especially in trusted information sources like Wikipedia.
Reliable detection of machine-generated text is critical for maintaining the integrity of online information and ensuring the quality and trustworthiness of platforms like Wikipedia, which influences public understanding.
This new benchmark (WETBench) specifically targets the detection of task-specific machine-generated text relevant to Wikipedia editing, which will lead to more effective and generalizable MGT detectors in real-world contexts.
- · Wikipedia editors
- · Integrity-focused AI research
- · Platforms dependent on factual accuracy
- · Malicious MGT actors
- · Low-quality LLM-generated content
Improved detection capabilities will help Wikipedia maintain its reputation as a reliable information source.
The development of robust MGT detection tools could lead to advancements in AI systems designed to circumvent such detectors, creating an arms race.
Enhanced trust in human-curated platforms could marginally shift user engagement away from less reliable, AI-saturated information sources.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL