Summary of How Well Do Llms Cite Relevant Medical References? An Evaluation Framework and Analyses, by Kevin Wu et al.
How well do LLMs cite relevant medical references? An evaluation framework and analyses
by Kevin Wu, Eric Wu, Ally Cassasola, Angela Zhang, Kevin Wei, Teresa Nguyen, Sith Riantawan, Patricia Shi Riantawan, Daniel E. Ho, James Zou
First submitted to arxiv on: 3 Feb 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 Large language models (LLMs) are revolutionizing medical question-answering across various clinical domains. Top-performing commercial LLMs can even cite sources to support their responses. However, do these generated sources actually support the claims made? We investigate this by proposing three contributions: first, we demonstrate that GPT-4 is highly accurate in validating source relevance; second, we develop an automated pipeline called SourceCheckup and evaluate five top-performing LLMs on a dataset of 1200 generated questions; and third, we open-source our curated dataset of medical questions and expert annotations. Our findings show that between 50% to 90% of LLM responses are not fully supported by their provided sources. This highlights the importance of understanding and quantifying LLM capabilities in producing trustworthy medical references. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can answer medical questions, but do they really support what they say? Some top models even give you a list of sources to back up their answers. We looked into this by making three contributions: first, we showed that GPT-4 is good at checking if a source is relevant; second, we made an automated tool called SourceCheckup and tested five top language models on 1200 medical questions; and third, we shared our dataset of questions and expert answers so others can test their own models. What we found was surprising: most of the time, these language models don’t fully support what they say with real sources. |
Keywords
» Artificial intelligence » Gpt » Question answering
