Equitas: A Corruption-Robustness Benchmark for Multi-LLM Committees
Overview
Equitas is a benchmark for evaluating aggregation strategies in hierarchical multi-LLM committees under adversarial corruption. It measures how well different aggregation methods maintain utility (task performance) and fairness (equitable outcomes across stakeholder groups) when a fraction of committee members are corrupted by adversaries.
All experiments use gpt-4o-mini as the underlying LLM… See the full description on the dataset page: https://huggingface.co/datasets/akshan-main/Equitas.
