<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>中国AI研究 on AI内参</title>
    <link>https://www.neican.ai/tags/%E4%B8%AD%E5%9B%BDai%E7%A0%94%E7%A9%B6/</link>
    <description>Recent content in 中国AI研究 on AI内参</description>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Thu, 31 Jul 2025 12:40:04 +0800</lastBuildDate>
    <atom:link href="https://www.neican.ai/tags/%E4%B8%AD%E5%9B%BDai%E7%A0%94%E7%A9%B6/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>ACL 2025前瞻：大模型深层“惯性”与效率重塑，AI安全与全球版图的交织</title>
      <link>https://www.neican.ai/insights/acl-2025ai-20250731124004558-0/</link>
      <pubDate>Thu, 31 Jul 2025 12:40:04 +0800</pubDate>
      <guid>https://www.neican.ai/insights/acl-2025ai-20250731124004558-0/</guid>
      <description>ACL 2025揭示了大模型发展中的两股核心力量：一方面是效率与长上下文能力的极致追求，以DeepSeek的硬件对齐稀疏注意力为代表；另一方面则是深层模型“抗对齐”特性的理论突破，北大团队的研究揭示了其内置“弹性”对齐难度，对AI安全与治理提出了严峻挑战。本届大会同时见证了中国科研力量的显著崛起，预示着全球AI研发格局的加速重塑。</description>
    </item>
  </channel>
</rss>
