Trevor Chan — LLM Ensembles

Overview

Despite huge advances, LLMs still lack convenient and reliable methods to quantify the uncertainty in their responses, making them difficult to trust in high-stakes applications. One of the simplest approaches to eliciting more accurate answers is to select the mode of many responses, a technique known as ensembling. In this work, we expand on typical ensembling approaches by looking at ensembles with a variable voting threshold. We introduce a theoretical framework for question answering and show that, by permitting ensembles to “abstain” from providing an answer when the dominant response falls short of the threshold, it is possible to dramatically increase the trustworthiness of the remaining answers. From this framework, we derive theoretical results as well as report experimental results on two problem domains: arithmetic problem solving and clinical-note question-answering. In both domains, we observe that large gains in answer trustworthiness can be achieved using highly restrictive voting ensembles, while incurring relatively modest reductions in response yield and accuracy. Due to this quality, voting ensembles may be particularly useful in applications - such as healthcare and data annotation - that require a high degree of certainty but which may not require that every question receive an automated answer.

Nair-Kanneganti, A., Chan, T.J., Goldfinger, S., Mackay, E., Anthony, B. and Pouch, A., 2025. Increasing LLM response trustworthiness using voting ensembles. arXiv preprint arXiv:2510.04048.

Simulated voting scenarios show that for a single question, deceptiveness δ alone dictates the ensemble accuracy as ensemble size approaches infinity. However, the rate of convergence is governed by the bewilderment.

Question difficulty is a function of deceptiveness (δ), a question's tendency to elicit a single specious answer, and bewilderment (η), the degree to which it encourages random guessing. (Illustration generated by ChatGPT.)

An ensemble of 50 models answers multiplication questions of varying difficulty. Notably, trust improves at high voting thresholds.

$A single model and ensemble extract salient features from the text of echocardiogram reports. Accuracy, yield, and trust are shown as a function of voting threshold and question distributions for model-extracted left ventricular ejection fraction.$

A single model and ensemble extract salient features from the text of echocardiogram reports. Accuracy, yield, and trust are shown as a function of voting threshold and question distributions for model-extracted left ventricular ejection fraction.