English(EN) Watermark Basins: Black-Box Detection of Distilled LLMs

新的WM-Basin方法可可靠检测蒸馏式大语言模型

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-28 04:00

研究人员开发了一种名为水印盆地对齐（WM-Basin）的新方法，用于在黑盒环境中检测蒸馏式大语言模型（LLMs）。该技术识别蒸馏模型参数空间中一个持久的几何信号，该信号与标准的微调方向正交。WM-Basin通过其API查询可疑模型，并使用探针模型将生成的文本映射到梯度表示，然后测试与水印盆地的对齐情况。与现有的令牌级统计方法相比，即使在量化和剪枝等蒸馏后修改之后，该方法也能显著提高检测率。 AI

影响增强了检测模型滥用和保护专有LLM输出的能力。

排序理由该集群包含一篇学术论文，详细介绍了一种检测蒸馏式大语言模型的新方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 15 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇学术论文，详细介绍了一种检测蒸馏式大语言模型的新方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Hongyan Chang, Hady Elsahar, Tom Sander, Pierre Fernandez, Saeed Mahloujifar, Sylvestre-Alvise Rebuffi, Tom\'a\v{s} Sou\v{c}ek, Tuan Tran, Valeriu Lacatusu, Surya Parimi, Rashel Moritz, Vanessa Stark, Alexandre Mourachko · 2026-10-28 04:00

Watermark Basins: 蒸馏LLM的黑盒检测

arXiv:2610.37266v1 Announce Type: cross Abstract: Proprietary language models and their reasoning traces are vulnerable to API-based distillation, where an adversary trains a student model on the teacher model's outputs. To obtain evidence of distillation, model providers can wat…

本文如何被排名

报道来源 [1]

Watermark Basins: 蒸馏LLM的黑盒检测

相关实体

相关话题