Hard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2302.03668/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTHard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery — mentioned by thatjoeoverthrhttps://news.ycombinator.com/item?id=44932169urn:hackerpapers:mention:2302.03668:44932169Sun, 17 Aug 2025 15:09:56 GMT<p>You can definitely &quot;snap&quot; it to the nearest neighbour according to the vocabulary matrix, but this comes with loss, so the &quot;snapped&quot; token won&#x27;t behave the same. Not sure how it would score on benchmarks. I&#x27;m thinking about how to approach this and I found this relevant paper: https://arxiv.org/pdf/2302.03668 I&#x27;m hoping I can tie this back into prefix tokens.</p><p><a href="https://hackerpaper.kaipfister.de/p/2302.03668/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LGHard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery — mentioned by thatjoeoverthrhttps://news.ycombinator.com/item?id=44932148urn:hackerpapers:mention:2302.03668:44932148Sun, 17 Aug 2025 15:07:44 GMT<p>Looking into this, I found this 2023 paper: https://arxiv.org/pdf/2302.03668 I haven&#x27;t gone through it yet but it seems they get tokenizable prompts on an image model. I don&#x27;t understand how you can backdrop all the way to the token IDs but I hope reading this will enlighten me and it would be fun to combine it with prefix tuning!</p><p><a href="https://hackerpaper.kaipfister.de/p/2302.03668/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.LG