Summary of Improving Socratic Question Generation Using Data Augmentation and Preference Optimization, by Nischal Ashok Kumar et al.
Improving Socratic Question Generation using Data Augmentation and Preference Optimization
by Nischal Ashok Kumar, Andrew Lan
First submitted to arxiv on: 1 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Computers and Society (cs.CY); 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 The paper presents a novel approach to augment the Socratic method in education by leveraging large language models (LLMs) like LLama 2. The goal is to generate Socratic questions that encourage students to think independently without providing direct answers or irrelevant prompts. Existing methods for generating these questions can produce invalid outputs, such as revealing the solution or asking premature questions. To address this issue, the authors propose a data augmentation method and a direct preference optimization (DPO) approach to optimize LLMs like LLama 2 to prefer ground-truth questions over generated invalid ones. The proposed methods are evaluated on a Socratic questions dataset for student code debugging, demonstrating that DPO-optimized LLama 2 can effectively avoid generating invalid questions and outperform existing prompting methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper talks about using big language models to help teachers ask better questions in the classroom. The goal is to make students think critically without giving them the answers. Right now, machines can sometimes generate bad questions that give away the answer or are just confusing. To fix this problem, the authors came up with two new ideas: one to make existing datasets of questions more diverse and another to train big language models to prioritize good questions over bad ones. They tested their approach on a dataset of questions about code debugging for students and found it worked better than other methods. |
Keywords
* Artificial intelligence * Data augmentation * Llama * Optimization * Prompting
