Eliciting Informative Text Evaluations with Large Language Models

Eliciting Informative Text Evaluations with Large Language Models
Abstract: In a wide variety of contexts including peer grading, peer review, and crowd-sourcing (e.g. evaluating LLM outputs) we would like to design mechanisms which reward agents for producing high quality responses. Unfortunately, computing rewards by comparing to ground truth or gold standard is often cumbersome, costly, or impossible. Instead we would like to compare agent reports.
Peer prediction mechanisms motivate high-quality feedback with provable guarantees. However, current methods only apply to rather simple reports, like multiple-choice or scalar numbers. We aim to broaden these techniques to the larger domain of text-based reports, drawing on the recent developments in large language models. This vastly increases the applicability of peer prediction mechanisms as textual feedback is the norm in a large variety of feedback channels: peer reviews, e-commerce customer reviews, and comments on social media. We introduce two mechanisms, the Generative Peer Prediction Mechanism (GPPM) and the Generative Synopsis Peer Prediction Mechanism (GSPPM). These mechanisms utilize LLMs as predictors, mapping from one agent’s report to a prediction of her peer’s report. Theoretically, we show that when the LLM prediction is sufficiently accurate, our mechanisms can incentivize high effort and truth-telling as an (approximate) Bayesian Nash equilibrium. Empirically, we confirm the efficacy of our mechanisms through experiments conducted on two real datasets: the Yelp review dataset and the ICLR OpenReview dataset. We highlight the results that on the ICLR dataset, our mechanisms can differentiate three quality levels — human written reviews, GPT-4-generated reviews, and GPT-3.5-generated reviews in terms of expected scores. Additionally, GSPPM penalizes LLM-generated reviews more effectively than GPPM.
No background is required, and the talk will introduce relevant background of peer prediction mechanisms.   
Bio: Grant Schoenebeck is an associate professor at the University of Michigan in the School of Information.  His work spans diverse areas in theoretical computer science but has recently focused on combining ideas from theoretical computer science, machine learning, and economics (e.g game theory, mechanism design, and information design) to develop and analyze systems for eliciting and aggregating information from of diverse group of agents with varying information, interests, and abilities.