FAPO automates LLM pipeline optimization via iterative diagnosis and prompt-or-structure edits, beating GEPA baseline by +14.1 pp mean across 18 comparisons and +33.8 pp when structural changes occur.
Universal Adversarial Triggers for Attacking and Analyzing NLP
7 Pith papers cite this work, alongside 34 external citations. Polarity classification is still indexing.
7
Pith papers citing it
34
external citations · OpenAlex
representative citing papers
Schützen is a German-Bulgarian LLM safety dataset showing pronounced cross-language differences in model safety behavior.
Translating unsafe inputs to low-resource languages jailbreaks GPT-4 at rates on par with or exceeding state-of-the-art attacks.
Baseline defenses including perplexity-based detection, input preprocessing, and adversarial training offer partial robustness to text adversarial attacks on LLMs, with challenges arising from weak discrete optimizers.
The Flan Collection demonstrates that task balancing, data enrichment, and mixed prompt training are critical to effective instruction tuning, yielding stronger Flan-T5 models released publicly.
Combines POS filtering and perplexity loss to generate sensible universal adversarial triggers that drop SST sentiment accuracy to 0.04-0.12, with adversarial training raising it to 0.48.
citing papers explorer
-
FAPO: Fully Automated Prompt Optimization of Multi-Step LLM Pipelines
FAPO automates LLM pipeline optimization via iterative diagnosis and prompt-or-structure edits, beating GEPA baseline by +14.1 pp mean across 18 comparisons and +33.8 pp when structural changes occur.
-
Sch\"utzen: Evaluating LLM Safety in Bulgarian and German Contexts
Schützen is a German-Bulgarian LLM safety dataset showing pronounced cross-language differences in model safety behavior.
-
Low-Resource Languages Jailbreak GPT-4
Translating unsafe inputs to low-resource languages jailbreaks GPT-4 at rates on par with or exceeding state-of-the-art attacks.
-
Baseline Defenses for Adversarial Attacks Against Aligned Language Models
Baseline defenses including perplexity-based detection, input preprocessing, and adversarial training offer partial robustness to text adversarial attacks on LLMs, with challenges arising from weak discrete optimizers.
-
The Flan Collection: Designing Data and Methods for Effective Instruction Tuning
The Flan Collection demonstrates that task balancing, data enrichment, and mixed prompt training are critical to effective instruction tuning, yielding stronger Flan-T5 models released publicly.
-
Universal Adversarial Triggers
Combines POS filtering and perplexity loss to generate sensible universal adversarial triggers that drop SST sentiment accuracy to 0.04-0.12, with adversarial training raising it to 0.48.
- Human agency in initial human-AI proof formalization workflows