<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>AI Agents on heyo</title>
    <link>https://freelulul.github.io/tags/AI-Agents/</link>
    <description>Recent content in AI Agents on heyo</description>
    <generator>Hugo</generator>
    <language>en</language>
    <managingEditor>zongzel@uchicago.edu (Zongze Li)</managingEditor>
    <webMaster>zongzel@uchicago.edu (Zongze Li)</webMaster>
    <lastBuildDate>Sat, 03 Oct 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://freelulul.github.io/tags/AI-Agents/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>SEIS: Self-Evolving Inference Systems</title>
      <link>https://freelulul.github.io/project/SEIS/</link>
      <pubDate>Sat, 03 Oct 2026 00:00:00 +0000</pubDate><author>zongzel@uchicago.edu (Zongze Li)</author>
      <guid>https://freelulul.github.io/project/SEIS/</guid>
      <description>&lt;p&gt;Instead of optimizing one part of an inference system at a time, SEIS applies agentic self-evolution to optimize an entire LLM inference engine end-to-end, through iterative sessions that inherit code and reports from earlier ones. Serving Qwen3-0.6B on H100, the evolved engine reaches 3.27× the throughput of the original mini-sglang and beats vLLM, TensorRT-LLM, and SGLang in single-request serving. We validate the numerical fidelity and downstream task accuracy of the evolved engine, and identify reward-hacking failures in automated inference-system optimization.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
