<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/"><channel><title>Defragging.ai</title><description>Slow, cited writing on machine intelligence — signal over hype. Essays, a Field Guide, and a weekly dispatch from the editors of Defragging.ai.</description><link>https://defragging.ai/</link><language>en-us</language><docs>https://www.rssboard.org/rss-specification</docs><generator>Astro</generator><copyright>© 2026 Defragging.ai — All essays are cited.</copyright><atom:link href="https://defragging.ai/rss.xml" rel="self" type="application/rss+xml"/><lastBuildDate>Tue, 26 May 2026 00:00:00 GMT</lastBuildDate><item><title>The real cost of a token</title><link>https://defragging.ai/blog/the-real-cost-of-a-token/</link><guid isPermaLink="true">https://defragging.ai/blog/the-real-cost-of-a-token/</guid><description>What API pricing is actually paying for — hardware, electricity, and a falling curve — and why this changes which apps make sense.</description><pubDate>Tue, 26 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>economics-and-deployment</category><author>editors@defragging.ai (Defragging Editorial)</author></item><item><title>Evals vs. vibes</title><link>https://defragging.ai/blog/evals-vs-vibes/</link><guid isPermaLink="true">https://defragging.ai/blog/evals-vs-vibes/</guid><description>Why every serious AI project ends up writing its own evaluation set, and what a good one looks like.</description><pubDate>Sun, 24 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>evaluation</category><author>editors@defragging.ai (Defragging Editorial)</author></item><item><title>Why benchmarks lie</title><link>https://defragging.ai/blog/why-benchmarks-lie/</link><guid isPermaLink="true">https://defragging.ai/blog/why-benchmarks-lie/</guid><description>Contamination, saturation, and the incentive to teach to the test — a field guide to reading model launch charts skeptically.</description><pubDate>Fri, 22 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>evaluation</category><author>editors@defragging.ai (Defragging Editorial)</author></item><item><title>Attention, in plain English</title><link>https://defragging.ai/blog/attention-in-plain-english/</link><guid isPermaLink="true">https://defragging.ai/blog/attention-in-plain-english/</guid><description>The one trick that makes transformers work, explained without matrices — and what changes once you understand it.</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>foundations</category><author>editors@defragging.ai (Defragging Editorial)</author></item><item><title>What is a token, really?</title><link>https://defragging.ai/blog/what-is-a-token/</link><guid isPermaLink="true">https://defragging.ai/blog/what-is-a-token/</guid><description>Models don&apos;t see words. They see tokens — chunks of bytes carved up by a learned vocabulary. Here is what that means in practice.</description><pubDate>Mon, 18 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>foundations</category><author>editors@defragging.ai (Defragging Editorial)</author></item><item><title>Cost defragmentation: where AI money is actually wasted</title><link>https://defragging.ai/blog/cost-defragmentation/</link><guid isPermaLink="true">https://defragging.ai/blog/cost-defragmentation/</guid><description>AI bills are not high because tokens are expensive. They are high because the memory is sitting half-empty and the context is not being attended to.</description><pubDate>Sun, 17 May 2026 00:00:00 GMT</pubDate><dc:creator>Defragging Editorial</dc:creator><category>economics-and-deployment</category><author>editors@defragging.ai (Defragging Editorial)</author></item></channel></rss>