<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Thomas — research</title><description>Research notes and experiments by thomasisalwayslate.</description><link>https://thomasisalwayslate.github.io</link><language>en</language><item><title>I lost to GhidraMCP on cost. Then I tested again.</title><link>https://thomasisalwayslate.github.io/research/ollama</link><guid isPermaLink="true">https://thomasisalwayslate.github.io/research/ollama</guid><description>Two modified Ollama builds. A first-round loss, then 5.6% lower cost than GhidraMCP with similar median times.</description><pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate></item><item><title>I beat a 300,000-star GitHub repo with 7 tools and a hacked copy of Doom</title><link>https://thomasisalwayslate.github.io/research/same-answers-lower-cost</link><guid isPermaLink="true">https://thomasisalwayslate.github.io/research/same-answers-lower-cost</guid><description>On modified Doom: same answers as REA, 52% fewer tokens, 36% cheaper and 27% faster.</description><pubDate>Sun, 18 Oct 2026 00:00:00 GMT</pubDate></item><item><title>huhu: the paths a life can take</title><link>https://thomasisalwayslate.github.io/research/huhu</link><guid isPermaLink="true">https://thomasisalwayslate.github.io/research/huhu</guid><description>Why I am building huhu, an operating system for a human: keeping what happens, following how one part of a life affects another, and learning the chances of things from the record itself.</description><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate></item></channel></rss>