<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>Qwen3-4B + VASPGuard vs GPT-5.4: INCARBench (2026 Preprint)</title>
        <link>https://peertube.watch/videos/watch/3c803ec8-fb27-41eb-a772-ccdc89d7f9fe</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-incar This is a system-level comparison, not a model-only comparison. The Qwen3-4B system used task-specific fine-tuning and VASPGuard, while GPT-5.4 was evaluated without VASPGuard. The results concern INCAR file generation on the authors' 500-case benchmark, not general AI capability. The paper was first posted on August 5, 2026. It is a research preprint, not a newly published peer-reviewed study. In a preprint by Zhang et al., a fine-tuned Qwen3-4B plus a rule-based checker (VASPGuard) - together called INCAR-SLM - scored 89.88 on INCARBench (500 cases; writing INCAR settings files for the VASP materials simulator), vs 74.33 for GPT-5.4 (Table 2, INCAR Score out of 100). The catch: the checker did a lot of the work. Fine-tuning alone scored 75.87, barely above GPT-5.4; the untuned Qwen3-4B with the checker alone scored 82.36. GPT-5.4 was not given the checker. One narrow task on the authors' own benchmark; Qwen3-8B (89.72) scored about the same as the 4B. Source: Zhang, Li, Shao, Yang, Liu, Wang, 'Fine-Tuning Small Language Models for Reliable VASP INCAR Generation', arXiv:2608.05387v1 - PREPRINT (Aug 5, 2026), not yet peer reviewed. https://arxiv.org/abs/2608.05387 . Numbers from Table 2. Names are trademarks of their respective owners; Peace Call is not affiliated with or endorsed by them. Charts drawn by code from the source's numbers. Paper: https://arxiv.org/abs/2608.05387 (version v1, posted Aug 5, 2026). AI Disclosure: Mr. Peace is a fictional AI character using a synthetic voice. This video is AI-assisted and does not depict a real person speaking. Research Disclosure: Based on a research preprint that has not been peer-reviewed. Results apply to the reported benchmarks and experimental conditions, not overall model superiority. Trademark Disclosure: Names are trademarks of their respective owners. Peace Call is not affiliated with or endorsed by them.</description>
        <lastBuildDate>Sat, 07 Nov 2026 15:49:38 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>PeerTube - https://peertube.watch</generator>
        <image>
            <title>Qwen3-4B + VASPGuard vs GPT-5.4: INCARBench (2026 Preprint)</title>
            <url>https://peertube.watch/lazy-static/avatars/3b284627-b75f-409e-ac2e-10f4c2946f43.png</url>
            <link>https://peertube.watch/videos/watch/3c803ec8-fb27-41eb-a772-ccdc89d7f9fe</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=3c803ec8-fb27-41eb-a772-ccdc89d7f9fe" rel="self" type="application/rss+xml"/>
    </channel>
</rss>