<?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>Transformers on Zero to One AI</title><link>https://blog.crazyai.uk/categories/transformers/</link><description>Recent content in Transformers on Zero to One AI</description><generator>Hugo</generator><language>en</language><lastBuildDate>Thu, 20 Aug 2026 11:36:00 +0800</lastBuildDate><atom:link href="https://blog.crazyai.uk/categories/transformers/index.xml" rel="self" type="application/rss+xml"/><item><title>The Extra Code Is Not Another Brain</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-06-extra-code-is-not-another-brain/</link><pubDate>Thu, 20 Aug 2026 11:36:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-06-extra-code-is-not-another-brain/</guid><description>Production nanoGPT adds a factory around the same next-token spine—and the product is still a document continuer, not a chatbot.</description></item><item><title>Width Paid, Context Lied, Last Step Lost</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-05-width-paid-context-lied/</link><pubDate>Thu, 20 Aug 2026 11:35:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-05-width-paid-context-lied/</guid><description>Three controlled knobs produced one useful winner, one confounded story, and a checkpoint that was better before training ended.</description></item><item><title>A Live Wire Is Not a Language</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-04-live-wire-is-not-a-language/</link><pubDate>Thu, 20 Aug 2026 11:34:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-04-live-wire-is-not-a-language/</guid><description>The model could memorize one batch, lower validation loss, restore a checkpoint, and imitate a play—without learning honest English.</description></item><item><title>The Residual Stream Is the Model</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-03-residual-stream-is-the-model/</link><pubDate>Thu, 20 Aug 2026 11:33:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-03-residual-stream-is-the-model/</guid><description>I stopped picturing a Transformer as layers replacing one another. A better picture is one long highway with repeated reads and writes.</description></item><item><title>Four Ways to Average the Past</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-02-four-ways-to-average-the-past/</link><pubDate>Thu, 20 Aug 2026 11:32:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-02-four-ways-to-average-the-past/</guid><description>Q, K, and V stopped feeling magical when I rebuilt attention as four increasingly selective ways to mix the same past.</description></item><item><title>The Batch Is a Factory of Futures</title><link>https://blog.crazyai.uk/posts/transformer-field-notes-01-batch-factory-of-futures/</link><pubDate>Thu, 20 Aug 2026 11:31:00 +0800</pubDate><guid>https://blog.crazyai.uk/posts/transformer-field-notes-01-batch-factory-of-futures/</guid><description>I thought training a language model meant teaching it to write. The first batch showed me something much more mechanical—and more useful.</description></item></channel></rss>