<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Capabilities of GPT-4 on Medical Challenge Problems · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2303.13375/</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/2303.13375/feed.xml" rel="self" type="application/rss+xml" /><item><title>Capabilities of GPT-4 on Medical Challenge Problems — mentioned by famouswaffles</title><link>https://news.ycombinator.com/item?id=35644725</link><guid isPermaLink="false">urn:hackerpapers:mention:2303.13375:35644725</guid><pubDate>Thu, 20 Apr 2023 19:08:52 GMT</pubDate><description>&lt;p&gt;Base GPT-4 was highly calibrated. read open ai&amp;#x27;s technical paper. also, this paper on gpt-4 performance of medical challenge problems confirmed the high calibration for medicine https://arxiv.org/abs/2303.13375&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2303.13375/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item><item><title>Capabilities of GPT-4 on Medical Challenge Problems — mentioned by famouswaffles</title><link>https://news.ycombinator.com/item?id=35508072</link><guid isPermaLink="false">urn:hackerpapers:mention:2303.13375:35508072</guid><pubDate>Mon, 10 Apr 2023 01:04:08 GMT</pubDate><description>&lt;p&gt;Interesting tidbits - Not all these models are instruct aligned - Cohere&amp;#x27;s suite is free for personal use - anthropic uses reinforcement learning from ai feedback for instruct alignment, removing humans from the process almost entirely(humans still make the rules the language model checks the responses against). - open ai&amp;#x27;s models are pretty poorly calibrated compared to its peers. Calibration is how well the model&amp;#x27;s confidence solving a problem tracks with how it actually performs that problem. It seems to be a consequence of the way they are performing the instruct tuning (or maybe it&amp;#x27;s a…&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2303.13375/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item><item><title>Capabilities of GPT-4 on Medical Challenge Problems — mentioned by fernly</title><link>https://news.ycombinator.com/item?id=35320823</link><guid isPermaLink="false">urn:hackerpapers:mention:2303.13375:35320823</guid><pubDate>Sun, 26 Mar 2023 23:23:25 GMT</pubDate><description>&lt;p&gt;In the full paper[1] read the examples of GPT-4&amp;#x27;s actual responses, e.g. Figs 3.1 and 7.1 and forward. Very, very impressive. [1] https://arxiv.org/pdf/2303.13375.pdf&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2303.13375/"&gt;Paper and discussion history&lt;/a&gt;. A mention is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item><item><title>Capabilities of GPT-4 on Medical Challenge Problems — HN submission</title><link>https://news.ycombinator.com/item?id=35319778</link><guid isPermaLink="false">urn:hackerpapers:submission:2303.13375:35319778</guid><pubDate>Sun, 26 Mar 2023 21:23:02 GMT</pubDate><description>&lt;p&gt;Capabilities of GPT-4 on Medical Challenge Problems&lt;/p&gt;&lt;p&gt;&lt;/p&gt;&lt;p&gt;147 points; 135 discussion replies on HN. &lt;a href="https://hackerpaper.kaipfister.de/p/2303.13375/"&gt;Paper and source history&lt;/a&gt;. A submission is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item><item><title>Capabilities of GPT-4 on Medical Challenge Problems — HN submission</title><link>https://news.ycombinator.com/item?id=35285091</link><guid isPermaLink="false">urn:hackerpapers:submission:2303.13375:35285091</guid><pubDate>Fri, 24 Mar 2023 04:28:40 GMT</pubDate><description>&lt;p&gt;Capabilities of GPT-4 on Medical Challenge Problems&lt;/p&gt;&lt;p&gt;&lt;/p&gt;&lt;p&gt;2 points; 0 discussion replies on HN. &lt;a href="https://hackerpaper.kaipfister.de/p/2303.13375/"&gt;Paper and source history&lt;/a&gt;. A submission is not an endorsement.&lt;/p&gt;</description><category>cs.CL</category></item></channel></rss>
