<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>工业应用 on mayulu的AI笔记</title><link>https://mayulu.co/tags/%E5%B7%A5%E4%B8%9A%E5%BA%94%E7%94%A8/</link><description>Recent content in 工业应用 on mayulu的AI笔记</description><generator>Hugo</generator><language>zhcn</language><copyright>Copyright © 2025–2026 mayulu All Rights Reserved</copyright><lastBuildDate>Tue, 22 Sep 2026 10:48:06 +0800</lastBuildDate><atom:link href="https://mayulu.co/tags/%E5%B7%A5%E4%B8%9A%E5%BA%94%E7%94%A8/index.xml" rel="self" type="application/rss+xml"/><item><title>Jev模型技术的原理和启示</title><link>https://mayulu.co/posts/jev-what-how-why-and-beyond/</link><pubDate>Tue, 22 Sep 2026 10:48:06 +0800</pubDate><guid>https://mayulu.co/posts/jev-what-how-why-and-beyond/</guid><description>Jev（System One Model）并非传统LLM，它放弃文本生成，直接输出带校准概率的类型安全决策，解决自动化场景下大模型延迟高、成本贵、置信不可靠的痛点。本文从第一性原理拆解Jev范式，厘清它与BERT、生成式大模型的本质差异，提炼四条核心技术启发，并推导7个同级别的下一代模型赛道。文章探讨异构AI系统架构：System2大模型负责开放式推理创作，System1专用原语承担机器侧高速决策，预示大一统模型时代走向终结。</description></item></channel></rss>