<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Meta on Peng Tan's AI Blog</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/</link><description>一个关注 AI 各领域的专题博客</description><atom:link href="https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/index.xml" rel="self" type="application/rss+xml"/><item><title>Agent训练新范式：Agent Learning via Early Experience</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/agent_learn_by_early_experience/</link><pubDate>Tue, 14 Oct 2025 18:00:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/agent_learn_by_early_experience/</guid><description>传统AI训练像是把人类所有的知识都强行灌输AI，而Meta的最新论文《Agent Learning via Early Experience》为我们展示了一条训练AI智能体的新路径: 可扩展、无需奖励的实用范式，通过将智能体自身的行为和结果转化为强大的监督信号，显著提升了AI的性能、数据效率和泛化能力。</description></item><item><title>Meta颠覆代码AI：新模型CWM不只“读”代码，更能“执行”它</title><link>https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/cwm_code_world_model/</link><pubDate>Sun, 28 Sep 2025 21:19:00 +0800</pubDate><guid>https://77f3dc70.hobbytp-github-io.pages.dev/zh/meta/cwm_code_world_model/</guid><description>Meta发布的一项名为“代码世界模型”（Code World Model, CWM）的全新研究，正试图从根本上解决当前代码AI面临的一大痛点。</description></item></channel></rss>