<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Jyu&apos;Log</title><description>Jun Yu 的研究、产品与写作。</description><link>https://junyu.blog/</link><language>zh-CN</language><item><title>Weak-to-Strong Exploitation: Weak LLMs Orchestrating Aligned Models Toward Harmful Goals</title><link>https://junyu.blog/zh/posts/unaware/</link><guid isPermaLink="true">https://junyu.blog/zh/posts/unaware/</guid><description>弱 LLM 作为恶意编排者：威胁模型、三类机制、因果评测与 RL 后续研究。</description><pubDate>Tue, 01 Jul 2025 00:00:00 GMT</pubDate></item><item><title>Stable Signer: 平衡三方利益的图像溯源协议</title><link>https://junyu.blog/zh/posts/stable-signer/</link><guid isPermaLink="true">https://junyu.blog/zh/posts/stable-signer/</guid><description>一串消息如何改变生成模型的参数，让图像携带可提取的标识？更重要的是，谁决定如何嵌入、谁有权提取、谁负责判断？</description><pubDate>Sun, 01 Sep 2024 00:00:00 GMT</pubDate></item></channel></rss>