Research map / Large language models / Language, safety and interpretability
Copy is All You Need
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 |
|---|---|---|---|
| Generate continuations by retrieving and copying context-sensitive phrases rather than predicting only from a fixed token vocabulary | CoG · WikiText-103 test | MAUVE: 23.43 -> 26.14 | Helped |
| Generate with CoG using greedy search instead of Transformer with nucleus sampling | CoG vs. Transformer · WikiText-103 test | MAUVE: 23.43 -> 26.01 | Helped |
| Use Law-MT as the retrieval collection without domain-specific training | CoG · Law-MT | MAUVE: 26.85 -> 28.14 | Helped |
| Use Law-MT as the retrieval collection without domain-specific training | CoG · Law-MT human evaluation | Human preference (Better): 36% -> 52% | Helped |
| Replace WikiText-103 retrieval memory with the larger En-Wiki collection without further training | CoG · WikiText-103 test | MAUVE: 23.43 -> 26.97 | Helped |
| Increase the En-Wiki phrase-index proportion above 1% (ablation) | CoG · WikiText-103 test | MAUVE | Helped |
| Increase the En-Wiki phrase-index proportion while it remains below 1% (ablation) | CoG · WikiText-103 test | MAUVE | No clear effect |
| Use Transformer fine-tuning on En-Wiki | Transformer · WikiText-103 test (En-Wiki setting) | MAUVE: 23.43 -> 21.31 | Hurt |
| Use nucleus sampling instead of greedy decoding | CoG · WikiText-103 test | MAUVE: 26.01 -> 26.14 | Mixed |
Closest papers by meaning
- Retrieval is Accurate Generation (ICLR 2024 · similarity 0.90)
- A Contrastive Framework for Neural Text Generation (NeurIPS 2022 · similarity 0.88)
- Towards Verifiable Text Generation with Generative Agent (AAAI 2025 · similarity 0.87)
- Reflection-Window Decoding: Text Generation with Selective Refinement (ICML 2025 · similarity 0.86)
- Compressed and Smooth Latent Space for Text Diffusion Modeling (NeurIPS 2025 · similarity 0.86)
- Controlled Text Generation via Language Model Arithmetic (ICLR 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.