https://111qqz.com/tags/agent-safety/Recent content in Agent-Safety on 111qqz的小窝Hugo -- gohugo.iozh-cnhust.111qqz@gmail.com (111qqz)hust.111qqz@gmail.com (111qqz)© 2011-2026 111qqzTue, 20 Oct 2026 18:50:00 +0800https://111qqz.com/2026/10/ai-agent-safety-beyond-accuracy/Tue, 20 Oct 2026 18:50:00 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/2026/10/ai-agent-safety-beyond-accuracy/<h2 class="relative group">起因 <div id="起因" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%b5%b7%e5%9b%a0" aria-label="锚点">#</a> </span> </h2> <p>过去几篇笔记一直在 Lecture 6 的范围里打转。<a href="https://111qqz.com/2026/10/distribution-shift-and-extrapolation/" >Distribution Shift 那篇</a>让我意识到离线测试集的 98% 准确率没法平移到生产环境，<a href="https://111qqz.com/2026/10/algorithmic-bias-and-fairness/" >Algorithmic Bias 那篇</a>又告诉我总体准确率可以掩盖群体间的巨大差异，<a href="https://111qqz.com/2026/10/adversarial-attacks-fgsm/" >对抗攻击那篇</a>则让我看到泛化好的模型仍然可以被微小扰动骗过。</p>