Summary of Extensive Self-contrast Enables Feedback-free Language Model Alignment, by Xiao Liu et al.
Extensive Self-Contrast Enables Feedback-Free Language Model Alignment
by Xiao Liu, Xixuan Song, Yuxiao Dong, Jie Tang
First submitted to arxiv on: 31 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper introduces Self-Contrast, a novel approach to large language model (LLM) alignment that leverages self-generated negatives without relying on human feedback. By fine-tuning LLMs with supervised targets and harnessing pre-trained embeddings to filter diverse candidates, Self-Contrast can effectively approximate balanced positive and negative preference annotations. The authors demonstrate the effectiveness of Self-Contrast through direct preference optimization (DPO) experiments on three datasets, outperforming standard DPO training by large margins. As the number of self-generated negatives increases, performance improves. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding a way to make big language models work better without needing as much help from humans. The problem with current methods is that they need lots of expensive human feedback or rely on other machines to make decisions. Self-Contrast is a new approach that uses the language model itself to generate many different options and then filters them based on how similar they are to each other. This makes it possible to improve the language model without needing as much help from humans. |
Keywords
* Artificial intelligence * Alignment * Fine tuning * Language model * Large language model * Optimization * Supervised
