Trending storyPast story
LLM watermarking degrades AI agent performance and speed
1 report1 sourceupdated 3 days ago
What happened
Summary
Lasso Security 测了一下给大模型加水印对 AI 智能体的影响，结果不太乐观。在 BFCL V3 这个工具调用基准测试上，加水印后模型调用工具的准确率掉了 3 到 5 个百分点，响应延迟多了 11 秒，输出长度也膨胀了 20% 到 30%。文章没提具体测的是哪些模型和智能体框架，所以这些数字只能当个参考。简单说，加水印相当于给模型输出打上隐...
Coverage
Follow the reports to see the story from different sides.
Sep 26
- Hacker News front pagePickLLM watermarking degrades AI agent performance and speed
Lasso Security tested LLM watermarking on AI agents and found it hurts tool-calling accuracy by 3–5 points on BFCL V3, adds 11 seconds of latency, and increases output length by 20–30%. The post doesn't name the specific models or agent frameworks tested.