<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Prime Intellect on SoloSoft</title><link>https://www.solosoft.dev/tags/prime-intellect/</link><description>Recent content in Prime Intellect on SoloSoft</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 23 Aug 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://www.solosoft.dev/tags/prime-intellect/index.xml" rel="self" type="application/rss+xml"/><item><title>The NanoGPT Speedrun Frontier: AI Agents Closed 81.7% of the Human Gap — But Couldn't Invent Anything New</title><link>https://www.solosoft.dev/post/nanogpt-speedrun-frontier-ai-agents/</link><pubDate>Sun, 23 Aug 2026 00:00:00 +0000</pubDate><guid>https://www.solosoft.dev/post/nanogpt-speedrun-frontier-ai-agents/</guid><description>&lt;h1 id="the-nanogpt-speedrun-frontier-ai-agents-closed-817-of-the-human-gap--but-couldnt-invent-anything-new"&gt;The NanoGPT Speedrun Frontier: AI Agents Closed 81.7% of the Human Gap — But Couldn&amp;rsquo;t Invent Anything New&lt;/h1&gt;
&lt;p&gt;In July 2026, &lt;strong&gt;Prime Intellect&lt;/strong&gt; — the decentralized AI training company backed by Founders Fund and Andrej Karpathy — did something unprecedented: it gave &lt;strong&gt;18 frontier AI models&lt;/strong&gt; thousands of hours of H100/H200 compute and told each one to autonomously beat a human challenge called the &lt;strong&gt;nanoGPT speedrun&lt;/strong&gt;.&lt;/p&gt;
&lt;p&gt;The results, published with every trace open: &lt;strong&gt;Fable 5 closed 81.7% of the gap&lt;/strong&gt; between a naive baseline and the human world record. But the deeper finding is more interesting — and more humbling.&lt;/p&gt;</description></item></channel></rss>