<?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>Exp_041 on Local First AI</title><link>https://localfirstai.eu/tags/exp_041/</link><description>Recent content in Exp_041 on Local First AI</description><image><title>Local First AI</title><url>https://localfirstai.eu/images/og-default.png</url><link>https://localfirstai.eu/images/og-default.png</link></image><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 18 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://localfirstai.eu/tags/exp_041/index.xml" rel="self" type="application/rss+xml"/><item><title>A Local Decision Model Made Our 4B Refuse 92 % of Unanswerable Questions and None of the Answerable Ones</title><link>https://localfirstai.eu/posts/2026-10-18-answerability-gate/</link><pubDate>Sun, 18 Oct 2026 00:00:00 +0000</pubDate><guid>https://localfirstai.eu/posts/2026-10-18-answerability-gate/</guid><description>Our 4B property model answers even when the retrieved text lacks the rule. A separate local decision model now checks first. The first test failed on wrong refusals, and every one of them was retrieval losing the source text. We fixed retrieval and tested again. On 150 sealed questions, Clef 27B raised correct refusals from 47 % to 92 %, refused none of the 101 answerable questions, and raised correctness from 75 % to 83 %.</description></item></channel></rss>