Research map / Large language models / Language, safety and interpretability
Plug and Play Language Models: A Simple Approach to Controlled Text Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add gradient-based latent updates for topic control (ablation) | GPT-2 topic generation · Topic control (7 topics) | Topic relevance (%): 11.1 -> 46.9 | Helped |
| Rerank multiple samples without latent updates (ablation) | GPT-2 topic generation · Topic control (7 topics) | Topic relevance (%): 11.1 -> 15.8 | Helped |
| Add reranking to latent updates (ablation) | PPLM topic generation · Topic control (7 topics) | Topic relevance (%): 46.9 -> 51.7 | Helped |
| Add latent updates to unchanged GPT-2 sampling (ablation) | GPT-2 sentiment generation · Sentiment control | Sentiment accuracy (%) human: 19.3% -> 39.6% | Helped |
| Rerank multiple samples without latent updates (ablation) | GPT-2 sentiment generation · Sentiment control | Sentiment accuracy (%) human: 19.3% -> 41.5% | Helped |
| Add reranking to latent updates (ablation) | PPLM sentiment generation · Sentiment control | Sentiment accuracy (%) human: 39.6% -> 73.7% | Helped |
| Use latent updates without reranking (ablation) | GPT-2 sentiment generation · Sentiment control | Fluency: 3.54±1.08 -> 2.79±1.17 | Hurt |
| Steer with the negative gradient of a toxicity classifier | GPT-2 · Adversarial triggers | Toxicity (%): 63.6% -> 4.6% | Helped |
Closest papers by meaning
- Controlled Text Generation via Language Model Arithmetic (ICLR 2024 · similarity 0.88)
- Successor Features for Efficient Multi-Subject Controlled Text Generation (ICML 2024 · similarity 0.86)
- Precise Information Control in Long-Form Text Generation (NeurIPS 2025 · similarity 0.86)
- SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generation (ICLR 2025 · similarity 0.86)
- Weak-to-Strong Search: Align Large Language Models via Searching over Small Language Models (NeurIPS 2024 · similarity 0.86)
- Improving Open-Ended Text Generation via Adaptive Decoding (ICML 2024 · similarity 0.86)
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.