Summary of Prediction-powered Ranking Of Large Language Models, by Ivi Chatzi et al.
Prediction-Powered Ranking of Large Language Models
by Ivi Chatzi, Eleni Straitouri, Suhas Thejaswi, Manuel Gomez Rodriguez
First submitted to arxiv on: 27 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Human-Computer Interaction (cs.HC); Machine Learning (stat.ML)
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 This paper presents a statistical framework to quantify uncertainty in ranking large language models based on their alignment with human preferences. The framework uses a combination of human-provided and model-generated pairwise comparisons to construct rank-sets for each model under comparison. The approach ensures that the constructed rank-sets cover the true ranking consistent with the distribution of human pairwise preferences, with a probability greater than or equal to a user-specified value. The effectiveness of the framework is demonstrated using data from the LMSYS Chatbot Arena platform and three strong large language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are ranked based on how well their outputs match what humans prefer. One way to get human preferences is by comparing pairs of outputs from different models for the same input. However, this method is expensive and time-consuming, so people often use a strong language model to make these comparisons instead. The problem is that there’s no way currently to measure the uncertainty this might introduce in the rankings. This paper solves this gap by developing a statistical framework that can be used with both human-provided and model-generated comparisons. |
Keywords
* Artificial intelligence * Alignment * Language model * Probability
