如你所见这只是一个日记本而已#赛博鸡蛋 一些羊毛#光与影 摄影作品或创作过程记录#吃点好 吃啥了#面基#活动#灵感菇 一些 MVP 的 demo 或 idea 记录点击讨论区进群友情链接：@stvgateway 联系Steven：@stvlynn_bothttps://diary.stv.pmhttps://diary.stv.pm/posts/4898https://diary.stv.pm/posts/4898Tue, 07 Apr 2026 15:20:02 GMT<a href="/search/%23%E4%BC%98%E8%B4%A8%E5%8D%9A%E6%96%87">#优质博文</a> <a href="/search/%23AI">#AI</a> <a href="/search/%23LLM">#LLM</a> <a href="/search/%23Context">#Context</a><br /><a href="https://www.developerway.com/posts/how-ai-remembers-and-forgets-part1" target="_blank">How AI Remembers and Why It Forgets: Part 1. The <mark>Context</mark> Problem</a>：介绍大语言模型（LLM）如何通过上下文（<mark>Context</mark>）模拟记忆，以及为何大量信息会导致“上下文腐烂”（<mark>Context</mark> Rot）现象。<br /><br /><blockquote>AI 摘要：文章揭示了 AI 并没有持久记忆，其所谓的“记忆”完全依赖于对话中反复发送的上下文（<mark>Context</mark>）。作者通过实验证明，即使在模型宣称的上下文窗口（<mark>Context</mark> Window）限制内，大量信息也会导致“上下文腐烂”（<mark>Context</mark> Rot），引发性能下降、信息遗漏及幻觉。</blockquote><br /><br /><i>author Nadia Makarevich</i><a href="https://www.developerway.com/posts/how-ai-remembers-and-forgets-part1" target="_blank"> <div>Developerway</div> <img class="link_preview_image" alt="How AI Remembers and Why It Forgets: Part 1. The Context Problem" src="/static/https://cdn4.telesco.pe/file/ZbcZcxMVE0_xRTeE6m8FnJ39vm3L162-SWekSg9wIWnjS3f2NHpU8rM0nZnCvIf2pmzzqo83dQPqE5xsuCqP6J8fLa-h5OGJRhmbTg_4UIg_FXVIuXv3X6dPAxQEFWkQ8KuEKbqKGfc8lXGk1l-xji8eFLhEzGmES3WrC3cqVhFaghQGMnMxuEZlplRBc0Ga0BABQgBVcj9vYzy_WUBQICspVwfNa-Yc0CILq5ivBgyVwiPbWuYW28zZwZlkslndkz0yJ3BgXDxwZq5-sMUi4ey-fvBT4kFOF-2_hd_gl9yfnaDZb4i3tfwOT2xsV7JfHaXcndVoD-pBfRYLGAK4sA.jpg" loading="lazy" /> <div>How AI Remembers and Why It Forgets: Part 1. The <mark>Context</mark> Problem</div> <div>How does AI actually remember things between messages, and why does it forget halfway through? I ran a few experiments on Claude Sonnet and GPT-5 and wrote down what I saw.</div> </a>