Summary of Towards Consistent Natural-language Explanations Via Explanation-consistency Finetuning, by Yanda Chen et al.
Towards Consistent Natural-Language Explanations via Explanation-Consistency Finetuning
by Yanda Chen, Chandan Singh, Xiaodong Liu, Simiao Zuo, Bin Yu, He He, Jianfeng Gao
First submitted to arxiv on: 25 Jan 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 The proposed explanation-consistency finetuning (EC-finetuning) method adapts large language models (LLMs) to generate consistent natural-language explanations on related examples. This is achieved by finetuning LLMs on synthetic data containing consistent explanations, which leads to a 10.0% relative improvement in explanation consistency across various question-answering datasets in different domains. The approach also generalizes well to out-of-distribution datasets not seen during training, with an additional +4.5% relative improvement. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) can create convincing and fluent explanations. However, they often provide inconsistent answers when asked similar questions. For instance, an LLM might say all birds can fly, but then answer no to the question about penguins flying. To help humans understand how these AI models make decisions, we want their answers to be consistent across related examples. A new method called explanation-consistency finetuning (EC-finetuning) helps AI models provide more consistent explanations. This is achieved by training the AI models on special data that has consistent answers. The results show that this approach makes AI models 10% better at providing consistent answers, and it also works well with datasets the AI model hasn’t seen before. |
Keywords
* Artificial intelligence * Question answering * Synthetic data
