Summary of What Large Language Models Know and What People Think They Know, by Mark Steyvers et al.
What Large Language Models Know and What People Think They Know
by Mark Steyvers, Heliodoro Tejeda, Aakriti Kumar, Catarina Belem, Sheer Karny, Xinyue Hu, Lukas Mayer, Padhraic Smyth
First submitted to arxiv on: 24 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Human-Computer Interaction (cs.HC)
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 As AI systems, particularly large language models (LLMs), become increasingly integrated into decision-making processes, the ability to trust their outputs is crucial. Recent work has focused on LLMs’ internal confidence, but less understood about how effectively they convey uncertainty to users. This paper explores the calibration gap and discrimination gap between human and model confidence in predicting answers. The authors conducted experiments with multiple-choice and short-answer questions, revealing that users tend to overestimate LLM response accuracy when provided with default explanations. Longer explanations increased user confidence, even if answer accuracy didn’t improve. By adjusting LLM explanations to reflect internal confidence, both gaps narrowed, significantly improving user perception of LLM accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary AI systems are getting better at helping us make decisions. But how do we know if the answers they give are correct? This paper looks at how well large language models (LLMs) can communicate uncertainty. The authors did some tests and found that people tend to think LLM answers are more accurate than they really are. They also found that giving people longer explanations helps them feel more confident, even if the answer isn’t actually better. By making sure LLMs explain themselves in a way that matches how confident they are, we can make people trust AI-assisted decision-making more. |
