<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Prefix-Tuning: Optimizing Continuous Prompts for Generation · HackerPapers discussions</title><link>https://hackerpaper.kaipfister.de/p/2101.00190/</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/2101.00190/feed.xml" rel="self" type="application/rss+xml" /><item><title>Prefix-Tuning: Optimizing Continuous Prompts for Generation — mentioned by gwern</title><link>https://news.ycombinator.com/item?id=26228076</link><guid isPermaLink="false">urn:hackerpapers:mention:2101.00190:26228076</guid><pubDate>Mon, 22 Feb 2021 18:18:18 GMT</pubDate><description>&lt;p&gt;https://arxiv.org/abs/2102.07350 already calls it &amp;quot;metaprompt&amp;quot; :) I gave it a quick stab a while ago but I think prompt programming is too new, and you can&amp;#x27;t easily cram demonstrations into an existing prompt, for it to really work well. It&amp;#x27;s more promising to train models on examples of tasks from instructions, or work on directly optimizing prompts for a goal (https://arxiv.org/abs/2101.00190) - it&amp;#x27;s a differentiable model and a whitebox, so use that power!&lt;/p&gt;&lt;p&gt;&lt;a href="https://hackerpaper.kaipfister.de/p/2101.00190/"&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>
