<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Activation Patching on 3rd layer</title><link>https://3rdlayer.uk/tags/activation-patching/</link><description>Recent content in Activation Patching on 3rd layer</description><generator>Hugo -- 0.154.5</generator><language>en-US</language><lastBuildDate>Sun, 16 Feb 2025 00:00:00 +0000</lastBuildDate><atom:link href="https://3rdlayer.uk/tags/activation-patching/index.xml" rel="self" type="application/rss+xml"/><item><title>Steering GPT-2's Emotions with Sparse Autoencoders</title><link>https://3rdlayer.uk/posts/sae-steering-gpt2/</link><pubDate>Sun, 16 Feb 2025 00:00:00 +0000</pubDate><guid>https://3rdlayer.uk/posts/sae-steering-gpt2/</guid><description>Finding emotion-related features in GPT-2 using OpenAI&amp;rsquo;s pretrained SAE, then training one from scratch. Feature patching turns &amp;lsquo;good person&amp;rsquo; into &amp;lsquo;shit&amp;rsquo;.</description></item></channel></rss>