<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>A Theoretical Analysis of the Repetition Problem in Text Generation · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2012.14660/</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/2012.14660/feed.xml" rel="self" type="application/rss+xml" /><item><title>A Theoretical Analysis of the Repetition Problem in Text Generation — mentioned by andreyk</title><link>https://news.ycombinator.com/item?id=29387740</link><guid isPermaLink="false">urn:hackerpapers:mention:2012.14660:29387740</guid><pubDate>Tue, 30 Nov 2021 01:38:42 GMT</pubDate><description>&lt;p&gt;This is a common problem with language models in general, and the reason is not that well understood. This paper A Theoretical Analysis of the Repetition Problem in Text Generation (https://arxiv.org/abs/2012.14660) seems to offer a principled answer. Basically the probability maximizing search procedure for text generation can get stuck in loops where the most likely next statement is similar or same to before. I&amp;#x27;m no NLP researcher so I don&amp;#x27;t have easy intuition on it, but that paper seems like a good read.&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2012.14660/"&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>
