Looking back at 2017's 'Generating Sentences by Editing Prototypes'
srush_nlp · x · 2026-08-26
Highlights the 2017 paper 'Generating Sentences by Editing Prototypes' by Kelvin Guu et al. The paper proposes a generative model that samples a prototype sentence from the training corpus and edits it into a new sentence. This approach improves perplexity and human evaluation scores compared to generation from scratch, yielding interpretable edit vectors.
More from Research
- LAION releases 10M-hour open video dataset BVD for multimodal training — HirokatuKataoka · 2026-08-27
- DAIR.AI indexes 3 years of papers with topic exploration and chat — omarsar0 · 2026-08-27
- AWS Reveals Advanced SFT Strategies: More Data Is Not Always Better — AWS ML Blog · 2026-08-27
- AWS Guide: High-Quality Data Beats Volume for SFT — AWS ML Blog · 2026-08-27
- Observers doubt Simile/Aaru claims over lack of datasets and peer review — daveholtz · 2026-08-27
- PrimeIntellect verifiers v0.3.1: Model Interception and Persistent ACP Sessions — xeophon · 2026-08-27