<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>144x Smaller AI Won... Until Math (0.5B vs 72B, Preprint)</title>
        <link>https://peertube.watch/videos/watch/f5c62b8c-55b4-4144-95bf-3757011b88a2</link>
        <description>Peace Call. Mr. Peace is an AI-made character; voice is AI-generated; script written with AI assistance. No affiliate links in this video. How this site is run: https://getpeacecall.com/verify/?utm_source=peertube&amp;utm_medium=video&amp;utm_campaign=underdog-ep3 Qwen2.5-0.5B beat Qwen2.5-72B (144x larger) on IMDB movie-review sentiment: 0.92 vs 0.89 accuracy. On GSM8K math the result flips: 72B 0.81 vs 0.5B 0.38. Source: Cao, Ma, Li, Ren, Chen, 'Task-Specific Efficiency Analysis: When Small Language Models Outperform Large Language Models', arXiv:2603.21389v1 - PREPRINT (Mar 22, 2026), not yet peer reviewed. https://arxiv.org/abs/2603.21389 . Accuracy from the paper's tables (IMDB: 0.92 vs 0.89; GSM8K: 0.81 vs 0.38). Model names are trademarks of their owners; not affiliated with Alibaba/Qwen or the authors. Charts drawn by code from the paper's numbers.</description>
        <lastBuildDate>Tue, 03 Nov 2026 13:47:35 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>PeerTube - https://peertube.watch</generator>
        <image>
            <title>144x Smaller AI Won... Until Math (0.5B vs 72B, Preprint)</title>
            <url>https://peertube.watch/lazy-static/avatars/3b284627-b75f-409e-ac2e-10f4c2946f43.png</url>
            <link>https://peertube.watch/videos/watch/f5c62b8c-55b4-4144-95bf-3757011b88a2</link>
        </image>
        <copyright>All rights reserved, unless otherwise specified in the terms specified at https://peertube.watch/about and potential licenses granted by each content's rightholder.</copyright>
        <atom:link href="https://peertube.watch/feeds/video-comments.xml?videoId=f5c62b8c-55b4-4144-95bf-3757011b88a2" rel="self" type="application/rss+xml"/>
    </channel>
</rss>