<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2402.12948/</link><description>Collected HN comments and submissions linking this paper.</description><language>en</language><lastBuildDate>Mon, 02 Nov 2026 11:16:22 GMT</lastBuildDate><atom:link href="https://hackerpaper.kaipfister.de/p/2402.12948/feed.xml" rel="self" type="application/rss+xml" /><item><title>GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick — mentioned by neuroticnews25</title><link>https://news.ycombinator.com/item?id=49328551</link><guid isPermaLink="false">urn:hackerpapers:mention:2402.12948:49328551</guid><pubDate>Mon, 17 Aug 2026 09:55:16 GMT</pubDate><description>&lt;p&gt;&amp;gt;It does not affect writing quality at all, provably Prove it, then? It&amp;#x27;s not a claim that GumbelSoft paper makes: &amp;quot;Regarding generation quality (perplexity), GumbelSoft shows relatively low perplexity&amp;quot; https://arxiv.org/html/2402.12948v3&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2402.12948/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item></channel></rss>
