Summary of Scaling Data Diversity For Fine-tuning Language Models in Human Alignment, by Feifan Song et al.
Scaling Data Diversity for Fine-Tuning Language Models in Human Alignment
by Feifan Song, Bowen Yu, Hao Lang, Haiyang Yu, Fei Huang, Houfeng Wang, Yongbin Li
First submitted to arxiv on: 17 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 investigates how to align large language models (LLMs) with human preferences without requiring excessive human feedback. The authors compare two approaches: increasing the diversity of prompts or responses, both of which require labeling. They find that fine-tuning LLMs with more diverse responses and fewer prompts is more effective in triggering human alignment. Additionally, they propose a new formulation for quantifying prompt diversity, which shows a linear correlation with the final performance of LLMs after fine-tuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores how to make large language models behave nicely and not generate mean or offensive content. The researchers compare two ways to help the model learn: giving it more diverse ideas to work from (prompts) or having it respond in different ways. They found that making the model respond in many different ways, but only getting a few prompts, works best. They also came up with a new way to measure how diverse the prompts are and showed that this diversity matters for the model’s performance. |
Keywords
» Artificial intelligence » Alignment » Fine tuning » Prompt
