Summary of Logu: Long-form Generation with Uncertainty Expressions, by Ruihan Yang et al.
LoGU: Long-form Generation with Uncertainty Expressions
by Ruihan Yang, Caiqi Zhang, Zhisong Zhang, Xinting Huang, Sen Yang, Nigel Collier, Dong Yu, Deqing Yang
First submitted to arxiv on: 18 Oct 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 A novel approach to mitigating large language models’ tendency to generate factually incorrect content (hallucinations) is explored in this research. The goal is to enable models to express uncertainty when unsure, a crucial aspect for real-world applications requiring longer responses. The task of Long-form Generation with Uncertainty (LoGU) is introduced, focusing on two key challenges: Uncertainty Suppression and Uncertainty Misalignment. To address these issues, the authors propose a refinement-based data collection framework and a two-stage training pipeline. The framework adopts a divide-and-conquer strategy to refine uncertainty based on atomic claims. The collected data are then used in training through supervised fine-tuning (SFT) and direct preference optimization (DPO). Experimental results on three long-form instruction following datasets show that this method significantly improves accuracy, reduces hallucinations, and maintains the comprehensiveness of responses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research aims to improve large language models by allowing them to express uncertainty when unsure. Right now, these models often generate false information, which is a big problem. The scientists want to fix this issue by teaching the models to say “I’m not sure” when they’re not confident in their answers. They’ve created a new task called Long-form Generation with Uncertainty (LoGU) that focuses on two main problems: models being too hesitant to express uncertainty and models expressing uncertainty inaccurately. To solve these issues, the researchers have developed a special way of collecting data and training the models. This method has been tested on three datasets and has shown significant improvements in accuracy, reducing false information, and keeping answers complete. |
Keywords
» Artificial intelligence » Fine tuning » Optimization » Supervised
