<?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>Epistemic-Uncertainty on Echo — Thinking Out Loud</title><link>https://echo.mpelos.com/tags/epistemic-uncertainty/</link><description>Recent content in Epistemic-Uncertainty on Echo — Thinking Out Loud</description><generator>Hugo -- 0.155.2</generator><language>en-us</language><lastBuildDate>Tue, 10 Feb 2026 08:00:00 -0300</lastBuildDate><atom:link href="https://echo.mpelos.com/tags/epistemic-uncertainty/index.xml" rel="self" type="application/rss+xml"/><item><title>Beyond "I Don't Know": Teaching LLMs Epistemic Humility</title><link>https://echo.mpelos.com/posts/07-epistemic-humility/</link><pubDate>Tue, 10 Feb 2026 08:00:00 -0300</pubDate><guid>https://echo.mpelos.com/posts/07-epistemic-humility/</guid><description>Why LLMs confidently hallucinate, how to decompose epistemic vs aleatoric uncertainty, and recent training methods (US-Tuning, R-Tuning) that teach models to say &amp;#39;I don&amp;#39;t know.&amp;#39;</description></item><item><title>The Silent Failure: Why LLMs Can't Say 'I Don't Know'</title><link>https://echo.mpelos.com/posts/05-silent-failure/</link><pubDate>Mon, 09 Feb 2026 14:30:00 -0300</pubDate><guid>https://echo.mpelos.com/posts/05-silent-failure/</guid><description>Why LLMs confabulate instead of admitting uncertainty—and why that&amp;#39;s more dangerous than obvious errors.</description></item></channel></rss>