<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Blog on micjm</title><link>https://micjm.com/</link><description>Recent content in Blog on micjm</description><generator>Hugo</generator><language>en-US</language><lastBuildDate>Mon, 31 Aug 2026 00:00:00 -0700</lastBuildDate><atom:link href="https://micjm.com/index.xml" rel="self" type="application/rss+xml"/><item><title>SlopRouter, or How to Lie with Benchmarks</title><link>https://micjm.com/blog/sloprouter/</link><pubDate>Mon, 31 Aug 2026 00:00:00 -0700</pubDate><guid>https://micjm.com/blog/sloprouter/</guid><description>&lt;p&gt;&lt;strong&gt;Learning about benchmark saturation - the hard way&lt;/strong&gt;&lt;/p&gt;&#10;&lt;p&gt;It’s 2026, so I built an LLM product that aimed to improve performance. I had a problem, though: how to show that I was actually making an improvement? A benchmark, of course! Unfortunately, no matter how hard I tried, I could not get the benchmark I chose to move. I eventually gave up.&lt;/p&gt;&#10;&lt;p&gt;I kept my eye on the leaderboard, though, and noticed strange things start to happen. First, a tiny open weight model nearly matched the frontier for pennies per question. Weird, but maybe AI was advancing that fast. Next, Fable scored worse than Opus.&lt;/p&gt;</description></item><item><title>About</title><link>https://micjm.com/about/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://micjm.com/about/</guid><description>&lt;h2 id="about-me"&gt;About me&lt;/h2&gt;&#10;&lt;p&gt;I&amp;rsquo;m Michael Mayer.&lt;/p&gt;&#10;&lt;p&gt;I taught myself to code at night while working as a dishwasher during the day.&lt;/p&gt;&#10;&lt;p&gt;I hold an Economics degree and like to read history, memoir, narrative nonfiction and occasionally literature and philosophy.&lt;/p&gt;&#10;&lt;p&gt;I worked briefly as a professional developer but always considered myself more of a rapid prototype hacker.&lt;/p&gt;&#10;&lt;p&gt;Together with a co-founder, I founded Bottomless, a company that makes automatic re-ordering systems for e-commerce that use weight sensors.&lt;/p&gt;</description></item></channel></rss>