<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Interpretable Traces, Unexpected Outcomes: Investigating the Disconnect in Trace-Based Knowledge Distillation · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2505.13792/</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/2505.13792/feed.xml" rel="self" type="application/rss+xml" /><item><title>Interpretable Traces, Unexpected Outcomes: Investigating the Disconnect in Trace-Based Knowledge Distillation — mentioned by internet_points</title><link>https://news.ycombinator.com/item?id=44245340</link><guid isPermaLink="false">urn:hackerpapers:mention:2505.13792:44245340</guid><pubDate>Wed, 11 Jun 2025 08:16:32 GMT</pubDate><description>&lt;p&gt;&amp;gt; I guess this means the reasoning traces are fully visible and not redacted in any way - interesting to see Mistral trying to turn that into a feature that&amp;#x27;s attractive to the business clients they are most interested in appealing to. but then someone found that, at least for distilled models, &amp;gt; correct traces do not necessarily imply that the model outputs the correct final solution. Similarly, we find a low correlation between correct final solutions and intermediate trace correctness https://arxiv.org/pdf/2505.13792 ie. the conclusion doesn&amp;#x27;t necessarily follow from the reasoning. So is…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2505.13792/"&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>
