<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Research on skulldc</title><link>https://skulldc.net/tags/research/</link><description>Recent content in Research on skulldc</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sat, 14 Nov 2026 10:26:06 +0000</lastBuildDate><atom:link href="https://skulldc.net/tags/research/index.xml" rel="self" type="application/rss+xml"/><item><title>PhysLoop uses LLM physics knowledge to steer video models toward self-improvement</title><link>https://skulldc.net/2026/11/physloop-uses-llm-physics-knowledge-to-steer-video-models-toward-self/</link><pubDate>Sat, 14 Nov 2026 10:26:06 +0000</pubDate><guid>https://skulldc.net/2026/11/physloop-uses-llm-physics-knowledge-to-steer-video-models-toward-self/</guid><description>PhysLoop wraps LLMs in a harness that guides video generation, selection and correction, and its authors release 100K videos plus two evaluation datasets.</description></item><item><title>SteerDistill lets a full-precision teacher correct a quantized agent while it works</title><link>https://skulldc.net/2026/11/steerdistill-lets-a-full-precision-teacher-correct-a-quantized-agent-w/</link><pubDate>Thu, 12 Nov 2026 10:25:20 +0000</pubDate><guid>https://skulldc.net/2026/11/steerdistill-lets-a-full-precision-teacher-correct-a-quantized-agent-w/</guid><description>SteerDistill is a quantization-aware distillation method in which a full-precision teacher steers a quantized student agent through its failed turns, and the authors report results within one point of BF16.</description></item><item><title>WikiTrace links each agent memory lesson back to the trajectory it came from</title><link>https://skulldc.net/2026/11/wikitrace-links-each-agent-memory-lesson-back-to-the-trajectory-it-cam/</link><pubDate>Thu, 12 Nov 2026 10:25:20 +0000</pubDate><guid>https://skulldc.net/2026/11/wikitrace-links-each-agent-memory-lesson-back-to-the-trajectory-it-cam/</guid><description>A KAIST AI paper argues that agent memory can make agents fail tasks they would otherwise solve, and proposes WikiTrace, which lets the agent check each stored lesson against its original trajectory.</description></item><item><title>Tree-Ising repairs a looping language model by flipping bits in its BF16 weights</title><link>https://skulldc.net/2026/11/tree-ising-repairs-a-looping-language-model-by-flipping-bits-in-its-bf/</link><pubDate>Wed, 11 Nov 2026 20:25:22 +0000</pubDate><guid>https://skulldc.net/2026/11/tree-ising-repairs-a-looping-language-model-by-flipping-bits-in-its-bf/</guid><description>Tree-Ising turns a gradient-free model repair into per-row QUBO problems, and the authors report it beats an LLM agent&amp;rsquo;s edit on MATH-500 using four forward-pass requests.</description></item></channel></rss>