Summary of Groundedness in Retrieval-augmented Long-form Generation: An Empirical Study, by Alessandro Stolfo
Groundedness in Retrieval-augmented Long-form Generation: An Empirical Study
by Alessandro Stolfo
First submitted to arxiv on: 10 Apr 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 empirical study investigates the phenomenon of groundedness in long-form question answering (LFQA) by retrieval-augmented large language models (LLMs). Specifically, it evaluates whether generated sentences are grounded in retrieved documents or pre-training data. The study finds that a significant fraction of generated sentences are ungrounded, even when containing correct ground-truth answers. Factors such as model size, decoding strategy, and instruction tuning also impact groundedness. Larger models tend to be more effective at grounding their outputs, but a substantial portion of correct answers remains compromised by hallucinations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary LFQA is like searching for answers online, but instead of using Google, large language models help find the best response. This study wants to know if these models always get it right and how they do it. They looked at three different datasets and four types of models, and found that many times, the models are actually guessing or making things up, even when their answers are correct! The size of the model seems to help, but there’s still a lot of room for improvement. |
Keywords
* Artificial intelligence * Grounding * Instruction tuning * Question answering
