WeWorkBuddy

Same Papers, Different Verdicts: Tracing Version Drift in LLM Peer Review

arXiv:2611.20608v1 Announce Type: new Abstract: Venues and benchmarks read LLM review scores as measurements of a paper, while providers replace their models within months. We audit twenty frontier models from OpenAI, Google and Anthropic on ICLR 2027 submissions while they were under review, with tests registered before the decisions. Between an earlier and a later version of Claude Opus and of Gemini, the share of papers placed on the accept side falls by 52 and 53 percentage points on 183 and

阅读原始来源
来源参考:Same Papers, Different Verdicts: Tracing Version Drift in LLM Peer Review。暂未取得截图。
来源记录 · 预览待采集 · arxiv.org打开来源查看完整网页。

来源记录

项目分析

暂无五维分析,可先查看上方来源资料。