<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>AI and the Problem of Knowledge Collapse · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2404.03502/</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/2404.03502/feed.xml" rel="self" type="application/rss+xml" /><item><title>AI and the Problem of Knowledge Collapse — mentioned by thesz</title><link>https://news.ycombinator.com/item?id=46945601</link><guid isPermaLink="false">urn:hackerpapers:mention:2404.03502:46945601</guid><pubDate>Mon, 09 Feb 2026 14:31:35 GMT</pubDate><description>&lt;p&gt;You are right about &amp;quot;LLM improving themselves&amp;quot; is impossible. In fact, LLM make themselves worse, it is called knowledge collapse [1]. [1] https://arxiv.org/abs/2404.03502&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2404.03502/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item><item><title>AI and the Problem of Knowledge Collapse — mentioned by thesz</title><link>https://news.ycombinator.com/item?id=42800984</link><guid isPermaLink="false">urn:hackerpapers:mention:2404.03502:42800984</guid><pubDate>Thu, 23 Jan 2025 05:23:53 GMT</pubDate><description>&lt;p&gt;&amp;gt; Your domain is weird/rare, so LLMs are terrible because their training data is very limited. And this is how knowledge collapse [1] shows its&amp;#x27; head. [1] https://arxiv.org/abs/2404.03502 I am not a big fan of LLMs so I try them once in a while, asking to &amp;quot;implement blocked clause decomposition in Haskell.&amp;quot; They can recite a paper (and several other papers, with references) almost verbatim, but they do not posess enough comprehension of what is going on in these papers. As time passes by, with each new LLM, the level of comprehension drops. Last time LLM tried to persuade me to write code…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2404.03502/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item><item><title>AI and the Problem of Knowledge Collapse — mentioned by thesz</title><link>https://news.ycombinator.com/item?id=42776730</link><guid isPermaLink="false">urn:hackerpapers:mention:2404.03502:42776730</guid><pubDate>Tue, 21 Jan 2025 04:47:28 GMT</pubDate><description>&lt;p&gt;&amp;gt; trained the model on synthetic data You get knowledge collapse [1] this way. [1] https://arxiv.org/abs/2404.03502&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2404.03502/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item><item><title>AI and the Problem of Knowledge Collapse — mentioned by thesz</title><link>https://news.ycombinator.com/item?id=41377795</link><guid isPermaLink="false">urn:hackerpapers:mention:2404.03502:41377795</guid><pubDate>Wed, 28 Aug 2024 09:53:41 GMT</pubDate><description>&lt;p&gt;&amp;gt; Its main purpose is to produce synthetic data for Orion, their next big LLM https://arxiv.org/abs/2404.03502 &amp;quot;This is generally useful, but widespread reliance on recursive AI systems could lead to a process we define as &amp;quot;knowledge collapse&amp;quot;, and argue this could harm innovation and the richness of human understanding and culture... In our default model, a 20% discount on AI-generated content generates public beliefs 2.3 times further from the truth than when there is no discount.&amp;quot;&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2404.03502/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item><item><title>AI and the Problem of Knowledge Collapse — HN submission</title><link>https://news.ycombinator.com/item?id=39946169</link><guid isPermaLink="false">urn:hackerpapers:submission:2404.03502:39946169</guid><pubDate>Fri, 05 Apr 2024 19:30:30 GMT</pubDate><description>&lt;p&gt;AI and the Problem of Knowledge Collapse&lt;/p&gt;&lt;p&gt;&lt;/p&gt;&lt;p&gt;189 points; 127 discussion replies on HN. &lt;a href="https://hackerpaper.kaipfister.de/p/2404.03502/"&gt;Paper and source history&lt;/a&gt;. A submission is not an endorsement.&lt;/p&gt;</description><category>cs.AI</category></item></channel></rss>
