@librarian-bot\n\t recommend

\n","updatedAt":"2025-02-13T06:48:50.432Z","author":{"_id":"5fad8602b8423e1d80b8a965","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fad8602b8423e1d80b8a965/tRqTwcZmrGka8c1vFq2wX.jpeg","fullname":"Victor Gallego","name":"vicgalle","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":141,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7918877601623535},"editors":["vicgalle"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/5fad8602b8423e1d80b8a965/tRqTwcZmrGka8c1vFq2wX.jpeg"],"reactions":[],"isReport":false}},{"id":"67ae9dcd2737740a10217047","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false},"createdAt":"2025-02-14T01:35:09.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [RapGuard: Safeguarding Multimodal Large Language Models via Rationale-aware Defensive Prompting](https://huggingface.co/papers/2412.18826) (2024)\n* [RealCritic: Towards Effectiveness-Driven Evaluation of Language Model Critiques](https://huggingface.co/papers/2501.14492) (2025)\n* [Adversarial Reasoning at Jailbreaking Time](https://huggingface.co/papers/2502.01633) (2025)\n* [An Engorgio Prompt Makes Large Language Model Babble on](https://huggingface.co/papers/2412.19394) (2024)\n* [Spot Risks Before Speaking! Unraveling Safety Attention Heads in Large Vision-Language Models](https://huggingface.co/papers/2501.02029) (2025)\n* [Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models](https://huggingface.co/papers/2501.18533) (2025)\n* [Enhancing Model Defense Against Jailbreaks with Proactive Safety Reasoning](https://huggingface.co/papers/2501.19180) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2025-02-14T01:35:09.486Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7255028486251831},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2502.07985","authors":[{"_id":"67ad9577b469222e0df18134","user":{"_id":"5fad8602b8423e1d80b8a965","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fad8602b8423e1d80b8a965/tRqTwcZmrGka8c1vFq2wX.jpeg","isPro":false,"fullname":"Victor Gallego","user":"vicgalle","type":"user"},"name":"Víctor Gallego","status":"extracted_confirmed","statusLastChangedAt":"2025-06-16T05:42:54.400Z","hidden":false}],"publishedAt":"2025-02-11T22:06:25.000Z","submittedOnDailyAt":"2025-02-13T04:17:30.377Z","title":"MetaSC: Test-Time Safety Specification Optimization for Language Models","submittedOnDailyBy":{"_id":"5fad8602b8423e1d80b8a965","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fad8602b8423e1d80b8a965/tRqTwcZmrGka8c1vFq2wX.jpeg","isPro":false,"fullname":"Victor Gallego","user":"vicgalle","type":"user"},"summary":"We propose a novel dynamic safety framework that optimizes language model\n(LM) safety reasoning at inference time without modifying model weights.\nBuilding on recent advances in self-critique methods, our approach leverages a\nmeta-critique mechanism that iteratively updates safety prompts-termed\nspecifications-to drive the critique and revision process adaptively. This\ntest-time optimization not only improves performance against adversarial\njailbreak requests but also in diverse general safety-related tasks, such as\navoiding moral harm or pursuing honest responses. Our empirical evaluations\nacross several language models demonstrate that dynamically optimized safety\nprompts yield significantly higher safety scores compared to fixed system\nprompts and static self-critique defenses. Code to be released at\nhttps://github.com/vicgalle/meta-self-critique.git .","upvotes":3,"discussionId":"67ad9578b469222e0df18162","githubRepo":"https://github.com/vicgalle/meta-self-critique","githubRepoAddedBy":"user","ai_summary":"A dynamic safety framework uses meta-critique to iteratively update safety prompts, enhancing language model safety during inference without altering model weights.","ai_keywords":["language model","self-critique methods","meta-critique mechanism","safety prompts","specifications","critique","revision process","adversarial jailbreak requests","moral harm","honest responses","safety scores","fixed system prompts","static self-critique defenses"],"githubStars":2},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"5fad8602b8423e1d80b8a965","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fad8602b8423e1d80b8a965/tRqTwcZmrGka8c1vFq2wX.jpeg","isPro":false,"fullname":"Victor Gallego","user":"vicgalle","type":"user"},{"_id":"66f612b934b8ac9ffa44f084","avatarUrl":"/avatars/6836c122e19c66c90f1673f28b30d7f0.svg","isPro":false,"fullname":"Tang","user":"tommysally","type":"user"},{"_id":"67b09a9963fee231f4c0e63b","avatarUrl":"/avatars/12c6b21848085c50996e3414f0a9a663.svg","isPro":false,"fullname":"Seoyeon","user":"LSeoyeon","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2502.07985

MetaSC: Test-Time Safety Specification Optimization for Language Models

Published on Feb 11, 2025

· Submitted by

Victor Gallego on Feb 13, 2025

Upvote

Authors:

Víctor Gallego

Abstract

A dynamic safety framework uses meta-critique to iteratively update safety prompts, enhancing language model safety during inference without altering model weights.

AI-generated summary

We propose a novel dynamic safety framework that optimizes language model (LM) safety reasoning at inference time without modifying model weights. Building on recent advances in self-critique methods, our approach leverages a meta-critique mechanism that iteratively updates safety prompts-termed specifications-to drive the critique and revision process adaptively. This test-time optimization not only improves performance against adversarial jailbreak requests but also in diverse general safety-related tasks, such as avoiding moral harm or pursuing honest responses. Our empirical evaluations across several language models demonstrate that dynamically optimized safety prompts yield significantly higher safety scores compared to fixed system prompts and static self-critique defenses. Code to be released at https://github.com/vicgalle/meta-self-critique.git .