<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Transformers represent belief state geometry in their residual stream · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2405.15943/</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/2405.15943/feed.xml" rel="self" type="application/rss+xml" /><item><title>Transformers represent belief state geometry in their residual stream — mentioned by hackinthebochs</title><link>https://news.ycombinator.com/item?id=44273887</link><guid isPermaLink="false">urn:hackerpapers:mention:2405.15943:44273887</guid><pubDate>Sat, 14 Jun 2025 02:54:37 GMT</pubDate><description>&lt;p&gt;LLMs are modelling the world, not just &amp;quot;predicting the next token&amp;quot;. They are not akin to &amp;quot;stochastic parrots&amp;quot;. Some examples here[1][2][3]. Anyone claiming otherwise at this point is not arguing in good faith. There are so many interesting things to say about LLMs, yet somehow the conversation about them is stuck in 2021. [1] https://arxiv.org/abs/2405.15943 [2] https://x.com/OwainEvans_UK/status/1894436637054214509 [3] https://www.anthropic.com/research/tracing-thoughts-language...&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2405.15943/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.LG</category></item><item><title>Transformers represent belief state geometry in their residual stream — mentioned by hackinthebochs</title><link>https://news.ycombinator.com/item?id=44224799</link><guid isPermaLink="false">urn:hackerpapers:mention:2405.15943:44224799</guid><pubDate>Mon, 09 Jun 2025 14:19:05 GMT</pubDate><description>&lt;p&gt;&amp;gt;But LLMs are fundamentally, completely, incapable of X. It is not something that can be a result of their processes. This is the very point of contention. You don&amp;#x27;t get to just assume it. &amp;gt; it is because it is, at its root, a statistical engine generating plausible next tokens, with no semantic understanding of the underlying data. Another highly contentious point you are just outright assuming. LLMs are modelling the world, not just &amp;quot;predicting the next token&amp;quot;. Some examples here[1][2][3]. Anyone claiming otherwise at this point is not arguing in good faith. It&amp;#x27;s interesting how the people…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2405.15943/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.LG</category></item><item><title>Transformers represent belief state geometry in their residual stream — mentioned by hackinthebochs</title><link>https://news.ycombinator.com/item?id=44092918</link><guid isPermaLink="false">urn:hackerpapers:mention:2405.15943:44092918</guid><pubDate>Mon, 26 May 2025 01:09:14 GMT</pubDate><description>&lt;p&gt;LLMs are modelling the world, not just &amp;quot;predicting the next token&amp;quot;. They are certainly not akin to parrots. Some examples here[1][2][3]. Anyone claiming otherwise at this point is not arguing in good faith. [1] https://arxiv.org/abs/2405.15943 [2] https://x.com/OwainEvans_UK/status/1894436637054214509 [3] https://www.anthropic.com/research/tracing-thoughts-language...&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2405.15943/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.LG</category></item></channel></rss>
