<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>超长上下文 on AI内参</title>
    <link>https://www.neican.ai/tags/%E8%B6%85%E9%95%BF%E4%B8%8A%E4%B8%8B%E6%96%87/</link>
    <description>Recent content in 超长上下文 on AI内参</description>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Mon, 08 Dec 2025 09:10:05 +0800</lastBuildDate>
    <atom:link href="https://www.neican.ai/tags/%E8%B6%85%E9%95%BF%E4%B8%8A%E4%B8%8B%E6%96%87/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>谷歌Titans与MIRAS：AGI纪元的新“基石”，重塑语言模型的记忆与未来</title>
      <link>https://www.neican.ai/insights/titansmirasagi-20251208091005022-1/</link>
      <pubDate>Mon, 08 Dec 2025 09:10:05 +0800</pubDate>
      <guid>https://www.neican.ai/insights/titansmirasagi-20251208091005022-1/</guid>
      <description>谷歌DeepMind推出了革命性的Titans架构和MIRAS理论框架，通过融合RNN与Transformer的优势，实现了高达200万token的超长上下文处理和“测试时训练”能力。这一突破不仅有效解决了传统Transformer的记忆瓶颈，更被视为加速AGI实现的关键一步，预示着AI在商业应用、多模态智能以及社会伦理层面将迎来深刻变革。</description>
    </item>
    <item>
      <title>类脑大模型“瞬悉”：中国GPU赋能下的AI范式重构与超长上下文潜能</title>
      <link>https://www.neican.ai/insights/gpuai-20250908201005701-0/</link>
      <pubDate>Mon, 08 Sep 2025 20:10:05 +0800</pubDate>
      <guid>https://www.neican.ai/insights/gpuai-20250908201005701-0/</guid>
      <description>中国科学院自动化研究所发布的“瞬悉1.0”类脑大模型，通过模仿大脑信息处理机制，实现了线性复杂度并全流程运行在国产GPU上。这一技术突破在超长序列处理上实现了百倍速度提升和显著能效优化，预示着AI架构可能从当前的Transformer模式转向“基于内生复杂性”的新范式，并为中国在AI核心技术领域实现自主可控提供了关键支撑。</description>
    </item>
  </channel>
</rss>
