Research map / Large language models / Language, safety and interpretability
Extensible Prompts for Language Models on Zero-shot Language Style Customization
Tao Ge, Jing Hu, Li Dong, Shaoguang Mao, Yan Xia, Xun Wang, Si-Qing Chen, Furu Wei
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove context-augmented learning (CAL) from X-Prompt (ablation) | X-Prompt · 20 Users (OOD) | perplexity: 28.5 -> 29.4 | Hurt |
| remove context-augmented learning (CAL) from X-Prompt (ablation) | X-Prompt · 20 Users (OOD) | accuracy: 38.6 -> 37.9 | Hurt |
| remove CAL from X-Prompt (ablation) | X-Prompt · 20 Users (ID) | perplexity: 30.8 -> 29.7 | Improved |
| remove CAL from X-Prompt (ablation) | X-Prompt · 800 Users (ID) | accuracy: 29.3 -> 29.9 | Improved |
| remove CAL from X-Prompt (ablation) | X-Prompt · Human evaluation, open-ended generation | overall: 0.54 -> 0.35 | Hurt |
| remove CAL from X-Prompt (ablation) | X-Prompt · Human evaluation, open-ended generation | content: 0.69 -> 0.38 | Hurt |
| use X-Prompt instead of prompt tuning | OPT-6.7b · 20 Users (OOD) | perplexity: 29.5 -> 28.5 | Improved |
| use X-Prompt instead of prompt tuning | OPT-6.7b · 20 Users (ID) | accuracy: 37.8 -> 37.2 | Hurt |
| increase few-shot examples from 8 to 32 | No prompt baseline · 800 Users (ID) | perplexity: 69.9 -> 62.7 | Improved |
| scale OPT from 350m to 6.7b | X-Prompt · Human evaluation, overall | overall: 0.24 (+0.09) -> 0.54 (+0.24) | Improved |
| scale OPT from 350m to 6.7b | X-Prompt · Human evaluation, content | content: 0.30 (+0.08) -> 0.69 (+0.35) | Improved |
| use X-Prompt instead of prompt tuning on style | OPT models · Human evaluation, style | style: 0.92 -> 0.83 | Hurt |
Closest papers by meaning
- Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting (ICLR 2024 · similarity 0.86)
- Evolving Prompts In-Context: An Open-ended, Self-replicating Perspective (ICML 2025 · similarity 0.86)
- Adaptive Prompt Routing for Arbitrary Text Style Transfer with Pre-trained Language Models (AAAI 2024 · similarity 0.86)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.85)
- In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering (ICML 2024 · similarity 0.85)
- Dialogue for Prompting: A Policy-Gradient-Based Discrete Prompt Generation for Few-Shot Learning (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
