<?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>AI Architecture on datafox.tw</title><link>https://datafox.tw/tags/ai-architecture/</link><description>Recent content in AI Architecture on datafox.tw</description><image><title>datafox.tw</title><url>https://datafox.tw/images/Open_graph_image.png</url><link>https://datafox.tw/images/Open_graph_image.png</link></image><generator>Hugo -- 0.146.0</generator><language>en</language><lastBuildDate>Thu, 14 May 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://datafox.tw/tags/ai-architecture/index.xml" rel="self" type="application/rss+xml"/><item><title>Why Modern LLMs Are Decoder-only: Architectural Evolution and Considerations from Seq2Seq to GPT</title><link>https://datafox.tw/posts/260514_1725/</link><pubDate>Thu, 14 May 2026 00:00:00 +0000</pubDate><guid>https://datafox.tw/posts/260514_1725/</guid><description>When Transformers first emerged, Seq2Seq seemed more logical. However, mainstream LLMs (e.g., GPT) are now exclusively Decoder-only. This article examines the architectural differences between Decoder and Seq2Seq, and their connection to Next Token Prediction, Scaling Laws, and emergent model capabilities.</description></item></channel></rss>