Nick Byrd, Ph.D.<p><a href="https://nerdculture.de/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> reasoning models may seem to reason reflectively when they say things like, "Let me rethink that".</p><p>But do these "reflective" phrases predict better reasoning performance?</p><p>Not in <a href="https://nerdculture.de/tags/Deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Deepseek</span></a> R1 Zero: <a href="https://doi.org/10.48550/arXiv.2503.20783" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">doi.org/10.48550/arXiv.2503.20</span><span class="invisible">783</span></a></p><p><a href="https://nerdculture.de/tags/cogSci" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>cogSci</span></a> <a href="https://nerdculture.de/tags/decisionScience" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>decisionScience</span></a> <a href="https://nerdculture.de/tags/processTracing" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>processTracing</span></a> <a href="https://nerdculture.de/tags/psychology" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>psychology</span></a></p>