Summary of To Know or Not to Know? Analyzing Self-consistency Of Large Language Models Under Ambiguity, by Anastasiia Sedova et al.
To Know or Not To Know? Analyzing Self-Consistency of Large Language Models under Ambiguity
by Anastasiia Sedova, Robert Litschko, Diego Frassinelli, Benjamin Roth, Barbara Plank
First submitted to arxiv on: 24 Jul 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 paper investigates the performance of large language models (LLMs) on applying factual knowledge when prompted with ambiguous entities. It proposes an evaluation protocol to disentangle knowing from applying knowledge and tests state-of-the-art LLMs on 49 ambiguous entities. The results show that LLMs struggle with choosing the correct entity reading, achieving an average accuracy of only 85%, and reveal systematic discrepancies in LLM behavior. This highlights the need to address entity ambiguity for more trustworthy LLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well big language models can use what they know when given ambiguous information. It makes a special test to see if the models are good at applying their knowledge, not just knowing it. The results show that these models aren’t very good at choosing the right answer from ambiguous information. They also find that the models have some biases and don’t always agree with each other. |
